Problem Statements Analysis

From what to build to how to build it — our platform helps you understand, plan, and solve SIH 2026 problem statements with complete insights into requirements, solution approaches, technology stacks, architecture, AI/ML, and implementation roadmaps.

Scroll down
01 / SIH 2026

PS: SIH26003

AI-Based Cognitive Gaming and Memory Assistance Platform for Elderly Dementia Patients in North Eastern Region (NER)

Description: Background: The North Eastern Region (NER) is witnessing a gradual rise in age-related cognitive disorders such as dementia and memory loss among the elderly population. Many families in remote and rural areas face challenges in accessing specialized neurological care, cognitive therapy, and long-term elderly support services due to limited healthcare infrastructure and geographical barriers. Elderly patients suffering from dementia often experience memory decline, confusion, anxiety, and social isolation, while caregivers face difficulties in continuous monitoring and engagement. There is limited availability of affordable and culturally inclusive digital therapeutic solutions tailored for elderly individuals in the North-Eastern Region. To strengthen elderly healthcare and improve cognitive well-being, there is a need for an accessible, engaging, and Al-enabled cognitive gaming solution designed specifically for dementia patients in NER. Description: This problem statement seeks the development of an AI-powered cognitive gaming and memory assistance platform for elderly dementia patients in the North Eastern Region. The solution should: a. Include interactive cognitive games and activities focused on: • Memory improvement • Attention and concentration 0 Daily routine recall • Pattern and object recognition of emotional and mental engagement b. Use AI/ML algorithms to adapt difficulty levels based on patient performance and cognitive condition c. Support multilingual and voice-assisted interaction suitable for elderly users in NER Include culturally familiar themes, visuals, sounds, and regional language support for d. better engagement e. Provide reminders for: • Medicines • Hydration • Daily activities • Medical appointments. f. Enable caregivers and healthcare workers to monitor patient progress through dashboards and activity levels g. Work in low-connectivity environments with offline functionality support h. Be accessible through mobile/tablet devices with a simple and elderly-friendly interface The platform should encourage long-term cognitive engagement, emotional well-being, and social interaction among elderly users. Expected Solution: A user-friendly Al-enabled cognitive assistance platform with: • Adaptive gaming and memory training modules • Voice-enabled multilingual interface • Cognitive performance tracking and analytics dashboard • Caregiver monitoring and alert system • Offline synchronization support for remote areas • Secure patient data management system • Simple and accessible UI/UX designed for elderly users The solution should support early cognitive intervention, improve quality of life for elderly dementia patients, and strengthen digital healthcare accessibility across the North Eastern Region.
Organization: Ministry of Development of North Eastern Region (MDoNER)
Department: Ministry of Development of North Eastern Region (MDoNER)
Category: Software
Theme: MedTech / BioTech / HealthTechtd>
Youtube Link: Not provided
Dataset Link: Not provided
Contact info: Not provided

1. Executive Snapshot

Summary

The elderly population in the North Eastern Region (NER) of India faces increasing rates of cognitive disorders such as dementia and Alzheimer's disease, compounded by severe geographical isolation, limited neurological care infrastructure, and significant linguistic diversity [PS, KNOWN]. Existing cognitive health platforms fail in this context because they require continuous high-speed cloud connectivity, rely on Western/urban motifs, and lack voice support for NER regional languages [INFERENCE]. This dossier outlines a local-first, AI-powered cognitive gaming and memory assistance platform named "AASHA-NER" (Adaptive Auxiliary & System for Health Assistance in NER) [ASSUMPTION]. Designed specifically for 4-6 student hackathon implementation, the platform combines on-device adaptive game difficulty using Item Response Theory (IRT) and Glicko-2 algorithms [RESEARCHED], voice-assisted interaction powered by the sovereign Digital India Bhashini AI stack [RESEARCHED], culturally familiar NER themes (e.g., Bihu motifs, Loktak lake flora, traditional weaving patterns) [RESEARCHED], local-first SQLite delta synchronization [RESEARCHED], and an integrated caregiver alert dashboard linked with primary healthcare workflows at Ayushman Arogya Mandirs and Tele-MANAS [RESEARCHED].

Metadata Table

Property Value Confidence Tag
Problem Statement ID SIH26003 [PS]
Title AI-Based Cognitive Gaming and Memory Assistance Platform for Elderly Dementia Patients in North Eastern Region (NER) [PS]
Organization Ministry of Development of North Eastern Region (MDoNER) [PS]
Department Ministry of Development of North Eastern Region (MDoNER) [PS]
Category Software [PS]
Theme MedTech / BioTech / HealthTech [PS]
Deadline 30 September 2026 [PS]
Ideas Counter 1/500 [PS]
Dataset Link None provided in PS (Requires open/synthetic neuropsychological data) [PS, INFERENCE]

Fundamental System Objective

To deliver a culturally tailored, voice-assisted, offline-first mobile cognitive stimulation and daily management platform that slows cognitive decline in elderly dementia patients across rural and remote North Eastern India while empowering family caregivers and community health workers with actionable analytics [INFERENCE].

2. Problem Deconstruction

Explicit Requirements & Source Text Extraction

The problem statement provided by MDoNER contains 8 explicit lettered requirements (a–h) and 7 core expected solution components [PS]:

Requirement a [PS]

  • Source Text: "Include interactive cognitive games and activities focused on: Memory improvement, Attention and concentration, Daily routine recall, Pattern and object recognition, Emotional and mental engagement."
  • Interpretation: The platform must provide at least five distinct mini-game/activity modules targeting these specific cognitive domains, designed specifically for elderly individuals experiencing mild-to-moderate cognitive decline [INFERENCE].
  • Implementation Implication: Game mechanics must avoid complex multi-finger gestures or fast-twitch time limits. Instead, they should use simple tap, drag, and voice inputs mapped to standardized neuropsychological screening frameworks like HMSE (Hindi Mental State Examination) and MUDRA (Multilingual Dementia Research and Assessment) [RESEARCHED].

Requirement b [PS]

  • Source Text: "Use AI/ML algorithms to adapt difficulty levels based on patient performance and cognitive condition."
  • Interpretation: Static level progression is unacceptable. The platform must dynamically adjust task complexity (number of items, visual noise, prompt speed, recall delay) in real time based on player error rate, response latency, and historical performance trends [INFERENCE].
  • Implementation Implication: An adaptive algorithm (such as 2-Parameter Logistic Item Response Theory or dynamic Glicko-2 rating) must run locally on the client device to recalculate difficulty parameters after every session without requiring cloud latency [RESEARCHED].

Requirement c [PS]

  • Source Text: "Support multilingual and voice-assisted interaction suitable for elderly users in NER."
  • Interpretation: Text-heavy interfaces fail with elderly dementia patients who may have reduced visual acuity or low literacy. Speech-to-Text (STT) and Text-to-Speech (TTS) must drive navigation and gameplay in major NER languages [INFERENCE].
  • Implementation Implication: Integration with Bhashini APIs (or offline pipeline fallbacks) for Assamese, Bengali, Manipuri (Meitei), Bodo, and Indian English voice models [RESEARCHED].

Requirement d [PS]

  • Source Text: "Include culturally familiar themes, visuals, sounds, and regional language support for better engagement."
  • Interpretation: Generic Western/urban game assets (e.g., skyscrapers, Western musical instruments) induce anxiety or confusion. Graphics, audio, and narratives must incorporate North Eastern cultural elements [INFERENCE].
  • Implementation Implication: Visual assets must feature regional themes such as Assam tea gardens, Majuli riverine motifs, Manipuri Pung Cholom rhythms, Meghalaya Living Root bridges, and traditional bamboo crafts [RESEARCHED]. Audio feedback should utilize authentic regional instruments (e.g., Pepa, Bahi/flute, Ektara) [RESEARCHED].

Requirement e [PS]

  • Source Text: "Provide reminders for: Medicines, Hydration, Daily activities, Medical appointments."
  • Interpretation: The app acts as a memory prosthetic for daily living activities (ADL). Reminders must be persistent, multi-modal (visual pop-ups, voice announcements in native language, physical vibration), and operable without internet access [INFERENCE].
  • Implementation Implication: Local alarm scheduling using native Android system alarms (AlarmManager / Exact Alarms) with audio chime and localized speech synthesis [INFERENCE].

Requirement f [PS]

  • Source Text: "Enable caregivers and healthcare workers to monitor patient progress through dashboards and activity levels."
  • Interpretation: A dual-interface system is required: a simplified UI for elderly patients and an analytical dashboard for family caregivers, Community Health Officers (CHOs), and ASHA workers at Ayushman Arogya Mandirs [RESEARCHED].
  • Implementation Implication: A secure caregiver web/mobile portal displaying key metrics: session frequency, domain-specific accuracy trends, task response times, adherence to medicine/hydration schedules, and behavioral anomaly flags [INFERENCE].

Requirement g [PS]

  • Source Text: "Work in low-connectivity environments with offline functionality support."
  • Interpretation: All core gameplay, adaptive logic, reminder scheduling, and local progress tracking must function 100% offline. Cloud communication should occur asynchronously via background delta synchronization when internet becomes available [INFERENCE].
  • Implementation Implication: Local-first database architecture (SQLite via WatermelonDB or RxDB) with a client-side sync queue, optimistic local updates, and conflict-free delta sync protocols [RESEARCHED].

Requirement h [PS]

  • Source Text: "Be accessible through mobile/tablet devices with a simple and elderly-friendly interface."
  • Interpretation: Touch targets must be extra-large (minimum 48x48 dp, ideally 64x64 dp), visual contrast must meet WCAG AAA standards, typography must be high-legibility, and navigation depth must not exceed 2 levels [INFERENCE].
  • Implementation Implication: Native Android/React Native/Flutter application optimized for low-cost Android tablets and smartphones (Android 8.0+, 2GB-4GB RAM) common in rural health settings [ASSUMPTION].

3. Stakeholders & User Needs

Stakeholder Goal Current Pain System Need
Elderly Dementia Patient (Early to Moderate Stage) [PS, INFERENCE] Maintain cognitive independence, feel emotionally supported, complete daily tasks safely, and play engaging games without frustration [INFERENCE]. Confusion, memory loss, inability to navigate complex phone screens, language barrier with English/Hindi apps, anxiety from high difficulty games [PS, KNOWN]. High-contrast, large-button voice UI; culturally familiar visual games; adaptive difficulty that prevents failure fatigue; native language audio reminders [PS, INFERENCE].
Family Caregiver (Son, Daughter, Spouse) [PS, INFERENCE] Ensure elderly relative takes medication on time, track cognitive status over time, reduce caregiving burnout, and receive early warning of sudden mental decline [INFERENCE]. Constant anxiety over unmonitored patient activities, lack of objective tracking tools, lack of specialized dementia care facilities in rural NER [PS, KNOWN]. Companion mobile/web dashboard; real-time miss-reminder alerts; visual progress graphs across memory/attention domains; SMS/WhatsApp notifications [PS, INFERENCE].
Community Health Worker / CHO / ASHA at Ayushman Arogya Mandir [RESEARCHED] Screen community elderly for cognitive decline, monitor registered dementia patients during monthly visits, and refer severe cases to tertiary care [RESEARCHED]. Lack of standardized digital screening tools in local languages, manual paper-based record keeping, inability to visit remote households frequently [RESEARCHED]. Multi-patient monitoring tablet UI; offline-first batch data sync; objective cognitive index scores (HMSE/MUDRA aligned) for clinical referral [RESEARCHED, INFERENCE].
Nodal Neurologist / Tele-MANAS Psychiatrist [RESEARCHED] Obtain long-term longitudinal data on patient cognitive trajectory to evaluate drug efficacy and therapeutic intervention [RESEARCHED]. Patient recall is highly unreliable; clinic visits occur only once every 6–12 months due to terrain and distance; complete absence of baseline cognitive logs [PS, KNOWN]. Exportable clinical dashboard (PDF/FHIR format) showing 90-day domain-specific cognitive trends, task latency distributions, and medication compliance rates [INFERENCE].

4. Requirements Traceability

Feature / Module Justifying Requirement Implementation Details
1. Memory Match Mini-Game ("Bihu Festival Items") Req a (Memory), Req d (Cultural motifs) Grid memory card matching using icons of traditional NER items (Gamocha, Jaapi, Pepa, Rhino, Kaziranga flora).
2. Pattern & Object Recognition Game ("Weaving Patterns") Req a (Pattern/Object), Req d Match traditional Muga silk / Handloom geometric weaving patterns under varying complexity.
3. Attention Focus Game ("Tea Leaf Picker") Req a (Attention/Concentration) Selective visual search task: tap target two-leaves-and-a-bud while ignoring distractor leaves.
4. Daily Routine Recall Game ("My Day Story") Req a (Daily Routine Recall) Sequential ordering of daily life tasks (brushing, morning tea, taking morning pill, lunch, evening walk).
5. On-Device IRT/Elo Adaptive Difficulty Engine Req b (AI/ML Adaptive Difficulty) Local C++/JS engine adjusting grid size, visual contrast, distractors, and response timer based on theta ($\theta$) score.
6. Bhashini Multilingual Speech Layer Req c (Multilingual & Voice) Voice navigation and spoken game prompts in Assamese, Bengali, Manipuri, Bodo, and Indian English.
7. Cultural Audio-Visual Asset Pack Req d (Cultural Familiarity) Folk instrument sound effects for correct/incorrect actions, calming regional background music, traditional visual themes.
8. Local Offline Alarm & Reminder Subsystem Req e (Reminders for Pills, Water, Appts) Native OS exact alarm triggers with multi-modal prompts (voice chime in native language, high-contrast modal screen).
9. Caregiver Analytics Portal & Mobile Dashboard Req f (Caregiver/Health Worker Monitoring) Web/App dashboard showing daily completion, cognitive performance radar chart, and missed reminder logs.
10. WatermelonDB / SQLite Local Sync Queue Req g (Low-Connectivity & Offline Support) Client-side SQLite database holding full app state; syncs encrypted JSON deltas when network returns.
11. Elderly-Centric High-Accessibility UI/UX Req h (Mobile/Tablet Elderly UI) Minimum 64px tap targets, high contrast (WCAG AAA), single-level hierarchy, physical back-button trap prevention.

5. Constraints & Environment Realities

Field Realities in North Eastern Region (NER)

  1. Geographical & Terrain Challenges [KNOWN]:

    • NER comprises 8 states characterized by hilly terrain, dense forests, and riverine islands (e.g., Majuli). Travel to district hospitals takes 4–12 hours.
    • System Constraint: The solution must not depend on physical clinical visits for regular operation.

  2. Connectivity & Power Infrastructure [KNOWN]:

    • Mobile network coverage in hill districts (e.g., Mon in Nagaland, Dima Hasao in Assam, East Khasi Hills in Meghalaya) fluctuates rapidly between 2G, 3G, and zero signal. Extended power outages (12–48 hours) are common during monsoon season.
    • System Constraint: The application must be 100% functional offline for up to 30 days. Data payloads during synchronization must be compressed to under 50 KB per sync cycle to transmit over weak 2G signals [INFERENCE].

  3. Linguistic Diversity & Literacy Levels [RESEARCHED]:

    • Over 220 languages/dialects are spoken across NER. Primary official regional languages include Assamese, Bengali, Manipuri (Meitei), Bodo, Mizo, Khasi, Garo, and Kokborok [KNOWN]. Literacy among rural elderly (ages 60+) is low, with high rates of illiteracy or inability to read fine smartphone text [KNOWN].
    • System Constraint: Text-only UIs will fail. The platform must be voice-first, icon-driven, and support major official regional languages (Assamese, Bengali, Manipuri, Bodo) with clear audio prompts [INFERENCE].

  4. Device Availability & Hardware Limits [ASSUMPTION]:

    • Rural households rely on budget Android smartphones or government-distributed tablets (e.g., under Ayushman Bharat / ASHA hardware schemes) with 2GB–4GB RAM, entry-level MediaTek/Snapdragon processors, and limited storage (32GB).
    • System Constraint: App binary size must stay under 40 MB. RAM usage must not exceed 150 MB. On-device AI processing must execute within < 50ms without overheating the device [INFERENCE].

  5. Physical & Cognitive Constraints of Dementia Patients [KNOWN]:

    • Dementia patients suffer from spatial disorientation, short-term memory failure, fine-motor tremors, visual contrast decline, and age-related presbycusis (hearing loss).
    • System Constraint: No multi-touch gestures (pinching, swiping with 3 fingers). Taps must have a minimum target diameter of 15mm. Audio frequency response must be tuned for clear voice reproduction [INFERENCE].

6. Research Questions

  1. Clinical Validation: What specific neuropsychological test items from ICMR MUDRA [RESEARCHED] and HMSE [RESEARCHED] can be accurately mapped to gamified mechanics without losing clinical correlation?
  2. On-Device Adaptive AI: How can Item Response Theory (IRT) or Elo/Glicko-2 algorithms be implemented in lightweight client-side JavaScript/Kotlin to run locally without server connectivity? [RESEARCHED]
  3. Voice AI Latency & Offline Fallback: How do Bhashini API web services perform under high-latency 2G conditions, and what lightweight offline TTS/STT fallback engines (e.g., Android Native TTS, Sherpa-ONNX, VOSK) exist for NER languages? [RESEARCHED]
  4. Cultural Resonance: What specific visual, auditory, and thematic assets induce positive emotional resonance and lower anxiety among elderly users across different NER sub-regions? [INFERENCE]
  5. Data Synchronization Safety: How can bidirectional synchronization handle extended offline periods (e.g., 30 days) with potential clock drift on low-end Android devices without corrupting patient cognitive history? [RESEARCHED]

7. Data Requirements & Sources

Data Type Purpose Ideal Real-World Source Hackathon Substitute Access Method Availability Privacy Sensitivity Confidence Tag
Patient Profile Demographics, primary language, baseline dementia stage (CDR/HMSE score). Electronic Health Records (EHR) via ABDM / ABHA ID system [RESEARCHED]. Local JSON profiles generated via on-boarding screen. Local SQLite / ABHA Sandbox API [RESEARCHED]. Requires ABHA Sandbox key [RESEARCHED]. High (PII / Health Data). [RESEARCHED]
Neuropsychological Game Item Bank Standardized cognitive tasks categorized by domain and difficulty score. ICMR MUDRA Toolbox / NIMHANS Neuropsychological battery [RESEARCHED]. Custom structured JSON item bank modeled on HMSE / MoCA sub-tests [RESEARCHED]. Embedded local JSON assets in app bundle. Openly designable based on published literature [RESEARCHED]. Low (Public Psychometric Formats). [RESEARCHED]
Game Session Logs Raw gameplay data: item ID, tap coordinates, response latency (ms), error count, hints used. Real-time sensor and touch events logged locally. Simulated gameplay event generators for testing + local touch event listeners. Local event logger into SQLite table. Immediately available on client. Medium (Behavioral Data). [INFERENCE]
Voice Audio Prompts Localized voice instructions for game rules, navigation, and reminders. Professional audio recordings by native speakers or Bhashini High-Quality TTS [RESEARCHED]. Synthetic voice audio generated via Bhashini TTS API [RESEARCHED] or cached MP3s. REST API call to Bhashini / Local MP3 assets [RESEARCHED]. Requires Bhashini API credentials [RESEARCHED]. Low (Public Audio). [RESEARCHED]
Medication & Daily Schedules Schedule times, dosage names, hydration frequency, doctor appointment dates. Caregiver input via companion app setup wizard. Pre-populated realistic sample schedule for demo profile. User form entry saved to SQLite. Available in app. High (Personal Health Data). [INFERENCE]
Aggregated Cognitive Metrics Calculated domain scores (Memory, Attention, Executive Function) over time. Computed on-device by IRT engine from session logs. Mathematical calculation script running on game session completion. Local database query / sync payload. Computed dynamically. High (Health Trend Data). [INFERENCE]

Data Classification Breakdown

  • A. Explicitly Mentioned in PS: Memory performance scores, daily routine logs, reminder schedules (medicines, hydration, activities, appointments), caregiver monitoring data [PS].
  • B. Derived by Inference: Detailed touch latency (ms), error retry counts, ambient connectivity logs, native language voice audio buffers, notification acknowledgment timestamps [INFERENCE].
  • C. To Be Collected During Usage: Patient touch interactions, voice command responses, medication adherence confirmations, session duration and frequency [INFERENCE].
  • D. Synthetically Generated for Hackathon: 30-day historical game session logs for 3 mock patients (Early Stage, Moderate Stage, Stable) to demonstrate caregiver dashboard trends [ASSUMPTION].
  • E. Data Needing Validation: Exact Bhashini voice model accuracy rates for specific NER tribal dialects (e.g., Kokborok, Garo) under noisy rural conditions [NEEDS VALIDATION].

8. Existing Solutions & Gap Analysis

Evaluated Systems

1. Global Cognitive Training Apps (CogniFit, Lumosity, Peak, BrainHQ) [RESEARCHED]

  • Description: Commercial mobile/web apps providing brain training games designed for general cognitive fitness or mild cognitive decline [RESEARCHED].
  • Strengths: Polished UX, large variety of games, scientifically structured difficulty scaling.
  • Weaknesses: Require continuous cloud internet; subscription costs ($10–$15/month); English/Western centered; zero cultural familiarity for NER rural elderly; no voice interaction in regional Indian languages; no integrated medication/hydration reminder engine; no connection to local public healthcare systems [INFERENCE].
  • Relevance: Benchmarks for game mechanics and cognitive domain targeting [INFERENCE].
  • Remaining Gap: Completely unsuited for rural, low-literacy NER elderly [INFERENCE].
  • Confidence Tag: [RESEARCHED]

2. ICMR MUDRA Toolbox (Multilingual Dementia Research & Assessment) [RESEARCHED]

  • Description: Neuropsychological assessment toolbox developed by ICMR consortium in 5 major Indian languages (Hindi, Bengali, Telugu, Kannada, Malayalam) for clinical diagnosis of dementia [RESEARCHED].
  • Strengths: Culturally validated for Indian populations; accounts for literacy levels; accepted as clinical standard [RESEARCHED].
  • Weaknesses: Diagnostic screening tool for trained clinicians, not a daily engaging therapeutic gaming system; does not offer offline mobile adaptive games or daily assistance reminders [INFERENCE].
  • Relevance: Provides the psychometric foundation for gamified task design [INFERENCE].
  • Remaining Gap: Needs translation into interactive, daily self-administered adaptive games with voice support in NER languages [INFERENCE].
  • Confidence Tag: [RESEARCHED]

3. Tele-MANAS & Ayushman Arogya Mandir (AAM) Workflows [RESEARCHED]

  • Description: Government of India’s 24/7 tele-mental health service (14416) integrated with primary care Community Health Officers (CHOs) at AAMs [RESEARCHED].
  • Strengths: Nationwide government infrastructure; supports 20+ regional languages; free 24/7 helpline; CHO/ASHA worker outreach [RESEARCHED].
  • Weaknesses: Currently tele-counseling and primary clinical screening focused; lacks an interactive continuous daily digital cognitive therapy and memory assistance companion app for patients at home [INFERENCE].
  • Relevance: The primary distribution, clinical referral, and monitoring backbone for production deployment [INFERENCE].
  • Remaining Gap: Digital patient-side daily engagement tool that feeds longitudinal data into Tele-MANAS / AAM clinician workflows [INFERENCE].
  • Confidence Tag: [RESEARCHED]

9. Candidate Solution Approaches

+---------------------------------------------------------------------------------------------------+
|                                CANDIDATE ARCHITECTURE COMPARISON                                  |
+------------------------------------+--------------------------------+-----------------------------+
| Approach 1: Pure Cloud Web App     | Approach 2: Monolithic App     | Approach 3: Hybrid Local    |
| (LLM + Cloud Backend)              | (Rule-Based Offline)           | (Local-First Adaptive +     |
|                                    |                                | Bhashini Voice)             |
| - High Cloud Dependency            | - Zero Cloud Dependency        | - Local-First SQLite        |
| - Severe Offline Latency Failures  | - Rigid Non-Adaptive Difficulty| - On-Device IRT Adaptive AI |
| - High Recurring API Costs         | - Poor Caregiver Analytics     | - Async Delta Cloud Sync    |
| - Poor Accessibility for Rural NER | - English/Hindi Text Only      | - Culturally Familiar UI    |
| [REJECTED]                         | [REJECTED]                     | [SELECTED]                  |
+------------------------------------+--------------------------------+-----------------------------+

Approach 1: Pure Cloud-Native Web Platform with Generative AI / LLM Integration

  • Mechanism: Next.js web application utilizing cloud LLM APIs (e.g., OpenAI / Claude) for dynamic dialogue, real-time cloud analytics, and server-side image generation for game levels.
  • Problem Fit: Low. Fails in rural NER due to zero/intermittent connectivity [INFERENCE].
  • Complexity: Medium (High backend complexity, low mobile client complexity).
  • Cost: High (Continuous per-API call costs for LLMs and cloud hosting).
  • Offline Feasibility: Zero (Application crashes without internet).
  • Verdict: REJECTED — Violates Requirement g (offline support) and introduces unpredictable operating costs [INFERENCE].

Approach 2: Monolithic Native App with Static Rule-Based Difficulty & Local Storage

  • Mechanism: Standalone Flutter/React Native Android app with pre-programmed level transitions (e.g., Level 1 -> Level 2 on 80% score), local SQLite storage, hardcoded text prompts in English/Hindi, and zero backend synchronization.
  • Problem Fit: Partial. Works offline, but fails on adaptive AI requirement, multilingual voice integration, and centralized caregiver/health worker monitoring [INFERENCE].
  • Complexity: Low.
  • Cost: Very Low.
  • Scalability: Extremely limited (Isolated silos of data, no caregiver dashboard).
  • Verdict: REJECTED — Fails Requirements b (AI/ML adaptive difficulty), c (voice multilingual), and f (caregiver dashboard) [INFERENCE].

Approach 3 (RECOMMENDED): Hybrid Local-First Mobile App with On-Device IRT Adaptive Engine, Bhashini Voice Layer, and Asynchronous Delta Sync

  • Mechanism:
    1. Client: React Native / Expo application running an embedded C++/JS 2-Parameter Item Response Theory (IRT) + Glicko-2 Adaptive Engine locally on device [RESEARCHED].
    2. Database: Local-first SQLite engine (via WatermelonDB) enabling sub-10ms local reads/writes and 100% offline functionality [RESEARCHED].
    3. Voice & Localization: Native Bhashini API integration for online STT/TTS with bundled local compressed audio fallbacks for offline operation [RESEARCHED].
    4. Backend: Node.js / NestJS server with PostgreSQL + Prisma for asynchronous delta synchronization when connectivity is detected [RESEARCHED].
    5. Caregiver Portal: Lightweight Next.js web dashboard for family members and ASHA/CHO health workers [INFERENCE].
  • Problem Fit: Excellent. Meets all requirements (a–h) while adhering to rural NER field constraints [INFERENCE].
  • Complexity: Medium-High (Requires clean offline sync queue and state machine logic).
  • Cost: Low (Open-source stack, Bhashini government sovereign AI, server needed only for lightweight sync).
  • Hackathon Feasibility: High (MVP can build 3 core mini-games, local IRT engine, local SQLite database, Bhashini voice integration, and caregiver web dashboard within available build window) [INFERENCE].
  • Verdict: SELECTED [INFERENCE].

10. Recommended Approach & Why

Why Approach 3 Wins

Approach 3 uniquely reconciles the conflicting demands of sophisticated AI adaptation and 100% offline rural usability [INFERENCE].

  1. Decoupled Intelligence: By bringing the adaptive AI algorithm (IRT/Glicko-2) into the client-side JavaScript bundle, difficulty tuning happens instantaneously after every game session without network requests [RESEARCHED].
  2. Local-First Reliability: The user experience is zero-latency because reads and writes happen against the embedded SQLite database [RESEARCHED]. Data loss is impossible even if the phone battery dies or the network drops mid-session [INFERENCE].
  3. Culturally & Linguistically Native: Integrating Bhashini directly into the UI architecture ensures that elderly users who cannot read can navigate via simple voice prompts in Assamese, Bengali, Manipuri, or Bodo [RESEARCHED].
  4. Clinical Ecosystem Integration: Asynchronous sync allows health workers at Ayushman Arogya Mandirs to download patient progress data during periodic home visits or when the patient enters a town with 4G coverage [RESEARCHED, INFERENCE].

11. Technical Architecture

Layered System Architecture Diagram

+---------------------------------------------------------------------------------------------------+
|                                      AASHA-NER SYSTEM ARCHITECTURE                                |
+---------------------------------------------------------------------------------------------------+

 [CLIENT LAYER - RURAL ELDERLY DEVICE (ANDROID TABLET / SMARTPHONE)]
 +-------------------------------------------------------------------------------------------------+
 |  Elderly-Friendly Voice & Touch UI (React Native / Expo - WCAG AAA)                             |
 |  +--------------------+  +--------------------+  +--------------------+  +-------------------+  |
 |  | Memory Improvement |  | Attention Focus    |  | Daily Routine Rec. |  | Pattern Match     |  |
 |  | Mini-Game          |  | Mini-Game          |  | Mini-Game          |  | Mini-Game         |  |
 |  +--------------------+  +--------------------+  +--------------------+  +-------------------+  |
 |  +-------------------------------------------------------------------------------------------+  |
 |  | Local Multi-modal Reminder Subsystem (Medicines, Hydration, Doctor Appts)                 |  |
 |  +-------------------------------------------------------------------------------------------+  |
 +-------------------------------------------------------------------------------------------------+
                                                |
                                                v
 +-------------------------------------------------------------------------------------------------+
 |  Embedded On-Device Core Engine (Runs 100% Offline)                                             |
 |  +-------------------------------------------------------------------------------------------+  |
 |  | On-Device Adaptive Engine: 2-Parameter IRT & Glicko-2 Difficulty Tuning Algorithm           |  |
 |  +-------------------------------------------------------------------------------------------+  |
 |  | Local Multi-Lingual Audio Manager (Bhashini Cached TTS / Android Native Speech)           |  |
 |  +-------------------------------------------------------------------------------------------+  |
 |  | Local Reactive Database: WatermelonDB on SQLite Engine                                    |  |
 |  | (Tables: Patients, GameSessions, CognitiveScores, Reminders, SyncQueue)                  |  |
 |  +-------------------------------------------------------------------------------------------+  |
 +-------------------------------------------------------------------------------------------------+
                                                |
                                                | (Background Delta Sync via Protobuf/HTTPS
                                                |  when 2G/3G/4G/Wi-Fi is available)
                                                v
 [CLOUD BACKEND LAYER (SECURE GOVT / CLOUD INFRASTRUCTURE)]
 +-------------------------------------------------------------------------------------------------+
 |  Sync & API Gateway (Node.js / Express / NestJS + REST / WebSocket)                             |
 |  +-------------------------------------------------------------------------------------------+  |
 |  | Delta Sync Engine: Conflict Resolution (Last-Write-Wins + Field Merge)                      |  |
 |  +-------------------------------------------------------------------------------------------+  |
 |  | Authentication & Security: JWT + HMAC Device Signature + AES-256 Storage                    |  |
 |  +-------------------------------------------------------------------------------------------+  |
 |  | Bhashini AI Wrapper Service (STT / TTS Proxy & Cache Manager)                             |  |
 |  +-------------------------------------------------------------------------------------------+  |
 |  | PostgreSQL Database + Prisma ORM (Centralized Analytics & Caregiver State)                |  |
 +-------------------------------------------------------------------------------------------------+
                                                |
                        +-----------------------+-----------------------+
                        |                                               |
                        v                                               v
 [EXTERNAL INTEGRATIONS]                         [PORTAL LAYER]
 +----------------------------------------+      +------------------------------------------------+
 | Digital India Bhashini APIs            |      | Caregiver & Health Worker Web Dashboard        |
 | (Assamese, Bengali, Manipuri, Bodo)    |      | (Next.js + Tailwind CSS + Recharts)            |
 | Tele-MANAS Helpline (14416) Integration|      | - Cognitive Radar Charts & Longitudinal Trends |
 | ABHA / ABDM Patient Health Record Sandbox|    | - Missed Medication & Hydration Alert Logs     |
 +----------------------------------------+      | - CHO / ASHA Worker Multi-Patient Overview     |
                                                 +------------------------------------------------+

Detailed Data-Flow Narrative

  1. User Interaction & Offline Session Start:

    • The elderly patient launches AASHA-NER on their Android device.
    • The app reads local patient profile and baseline theta score ($\theta$) from the on-device WatermelonDB (SQLite) in < 10ms [RESEARCHED].
    • The local UI announces the daily activity in the user's selected native language (e.g., Assamese) using cached audio files or local TTS synthesis [RESEARCHED].

  2. Gameplay & Adaptive Difficulty Adjustment:

    • The patient plays a cognitive game (e.g., Memory Match with Bihu items).
    • Touch events, response times (ms), hints requested, and errors are captured locally by the Game Session Logger [INFERENCE].
    • Upon game completion, the On-Device IRT Engine executes locally: $$\theta_{new} = \theta_{old} + K \cdot (Score_{observed} - Score_{expected}(\theta_{old}, b_i))$$ Where $b_i$ is the difficulty rating of item $i$. The updated ability parameter ($\theta_{new}$) immediately tunes the difficulty parameters for the next game session [RESEARCHED].

  3. Local Persistence & Reminder Scheduling:

    • The session result, score, latency distribution, and updated ability score are stored in local SQLite tables (GameSessions, CognitiveScores) and appended to the SyncQueue table [RESEARCHED].
    • Meanwhile, the Local Reminder Engine uses Android AlarmManager to fire scheduled exact alarms for medication or hydration, displaying full-screen high-contrast modal prompts with local audio announcements even if the main app is closed [INFERENCE].

  4. Background Asynchronous Delta Synchronization:

    • The Connectivity Listener continuously monitors network status using NetInfo [RESEARCHED].
    • When network connectivity (2G/3G/4G/Wi-Fi) is detected, a background sync worker pulls pending records from SyncQueue [RESEARCHED].
    • It sends a compressed JSON payload (Delta Sync) containing last_pulled_at timestamp and created/updated records to the Cloud Sync Gateway over TLS 1.3 [RESEARCHED].
    • The Cloud Sync Gateway validates the JWT token, applies field-level conflict resolution (Server-wins for caregiver overrides, Client-wins for game performance records), writes to PostgreSQL, and returns new server state (e.g., updated medication schedules set by the caregiver) [RESEARCHED, INFERENCE].
    • Upon receipt of HTTP 200 OK, the local SyncQueue records are marked as synced [RESEARCHED].

  5. Caregiver & Health Worker Dashboard View:

    • Family caregivers or ASHA workers log into the Next.js web portal.
    • The portal fetches synchronized analytics from PostgreSQL and renders domain-specific cognitive trajectory charts (Memory, Attention, Executive Function), adherence rates, and alert flags [INFERENCE].
    • If a patient misses 3 consecutive medication reminders or exhibits a >20% decline in attention scores over 7 days, an automated SMS/WhatsApp notification is triggered to the caregiver [INFERENCE].

12. Technology Stack

Layer Selected Technology Why Selected Alternative Evaluated Why Alternative Rejected
Mobile Frontend React Native + Expo (SDK 51+) Rapid cross-platform build, excellent native bridge performance, robust SQLite support, huge UI ecosystem for accessible design [RESEARCHED]. Flutter Slower C++ bindings for local WatermelonDB SQLite sync; higher learning curve for student team [INFERENCE].
Local Database Engine WatermelonDB (SQLite) High performance reactive database optimized for React Native; built on SQLite via JSI C++ bridge; lazy loading handles thousands of records smoothly; built-in delta sync protocol [RESEARCHED]. Realm / AsyncStore AsyncStore lacks relational queries and performance; Realm has proprietary engine licensing overhead and larger binary footprint [RESEARCHED].
Cloud Backend Framework Node.js (NestJS / Express) Asynchronous non-blocking I/O ideal for sync gateways; unified TypeScript codebase with React Native frontend; light footprint [INFERENCE]. Python Django / Fast API Higher memory consumption on cheap cloud instances; slower JSON serialization for bulk delta sync [INFERENCE].
Primary Database (Cloud) PostgreSQL (v16) + Prisma ORM Enterprise-grade relational integrity; JSONB column support for flexible game telemetry logs; Prisma provides type-safe DB access [INFERENCE]. MongoDB Document stores lack transactional integrity required for multi-patient health logs and schedule sync [INFERENCE].
Voice AI & Speech Engine Digital India Bhashini APIs + AI4Bharat Official sovereign AI stack of Govt. of India supporting 22 Indian scheduled languages including Assamese, Bengali, Manipuri, Bodo [RESEARCHED]. Google Cloud Speech / Whisper Lacks accuracy in regional NER dialects; high per-minute commercial costs; non-compliant with sovereign Indian data storage guidelines [RESEARCHED].
Caregiver Portal Frontend Next.js (v14 React Framework) + Tailwind CSS Server-side rendering for fast dashboard load; Tailwind for clean responsive UI; Recharts library for clear cognitive trend graphs [INFERENCE]. Angular Steep learning curve; verbose boilerplate unsuitable for rapid hackathon execution [INFERENCE].
State Management Zustand + React Query Lightweight, predictable state management without Redux boilerplate; React Query handles server state caching and optimistic UI updates seamlessly [INFERENCE]. Redux Toolkit Excessive boilerplate code; over-complicates simple mobile state flows [INFERENCE].
Hosting & Cloud Infrastructure Docker Containers on Render / Railway / AWS EC2 Simple containerized deployment; predictable cost tier; full control over SSL and environment variables [INFERENCE]. Serverless AWS Lambda Cold start latency degrades real-time sync response; difficult to run persistent local mock databases during SIH demo [INFERENCE].

13. AI/ML Justification

Module 1: On-Device Adaptive Difficulty Engine (Psychometric IRT + Glicko-2)

  1. What exact problem does AI solve? Prevents cognitive fatigue/frustration (if game is too hard) or boredom/disengagement (if game is too easy) by dynamically adjusting game parameters to match the patient's real-time cognitive capacity [PS, INFERENCE].
  2. What is the input? Game session telemetry: item ID, correct/incorrect status, completion time ($T_{obs}$ in ms), hints requested, current patient ability parameter ($\theta$).
  3. What is the output? Updated patient ability score ($\theta_{new}$) and item parameter selection (grid dimension $N \times M$, target delay time $\tau$, visual distractor count $D$) for the next session.
  4. Why can a simpler rule/system not solve it? Fixed rules (e.g., "if score > 80% go to Level 2") fail in dementia care because cognitive decline is non-linear and day-dependent (e.g., "sundowning" effects where cognitive function drops in late afternoon). IRT models the probability of a correct response as a continuous function of latent cognitive ability and item difficulty [RESEARCHED].
  5. What data is required? Pre-calibrated item difficulty parameters ($b_i$) and discrimination parameters ($a_i$) for each game task, derived from neuropsychological standards (HMSE/MUDRA) [RESEARCHED].
  6. How will it be evaluated? Target response success rate maintained within 65%–75% (the optimal flow state for cognitive rehabilitation without anxiety) [RESEARCHED].
  7. What happens when it is wrong? If difficulty spikes, the patient fails a task; the algorithm detects consecutive errors ($N \ge 2$) and drops difficulty by $2\sigma$ immediately [INFERENCE].
  8. Does it require training? No continuous deep learning training required. Uses standard psychometric parameter estimation formulas (Maximum Likelihood / Bayesian EAP) running locally [RESEARCHED].
  9. Fine-tuning? No.
  10. RAG? No.
  11. Computer Vision? No.
  12. Speech Processing? No.
  13. Forecasting? Yes, forecasts expected probability of success $P(Y=1|\theta, b_i)$.
  14. Human Oversight? Caregivers can view the estimated $\theta$ trajectory and manually cap maximum game difficulty if recommended by a physician [INFERENCE].

Module 2: Multilingual Voice Interaction Engine (Bhashini API Integration)

  1. What exact problem does AI solve? Enables illiterate or motor-impaired elderly users to navigate the app, receive spoken instructions, and respond to memory prompts using natural voice in regional languages [PS, RESEARCHED].
  2. What is the input? Spoken audio input from patient mic (for STT); text strings for game prompts/reminders (for TTS).
  3. What is the output? Recognized text intent / command (STT output); synthesized natural audio stream in native language (TTS output).
  4. Why can a simpler system not solve it? Rule-based audio playback only works for pre-recorded fixed clips. Dynamic game prompts, localized reminder announcements, and voice-command recognition require acoustic deep learning models trained on Indian regional accents [RESEARCHED].
  5. What data is required? Speech corpora for Assamese, Bengali, Manipuri, Bodo provided by Bhashini / AI4Bharat open datasets [RESEARCHED].
  6. How will it be evaluated? Word Error Rate (WER) < 15% for primary regional languages under clear mic conditions [RESEARCHED].
  7. What happens when it is wrong? The app displays large icon choices on screen alongside voice prompts and allows simple single-tap manual overrides [INFERENCE].
  8. Training? Pre-trained models hosted on Bhashini infrastructure [RESEARCHED].
  9. Fine-tuning? Not during hackathon.
  10. RAG? No.
  11. Computer Vision? No.
  12. Speech Processing? Yes (STT / TTS).
  13. Forecasting? No.
  14. Human Oversight? Caregivers can select and verify the preferred dialect/accent in settings [INFERENCE].

Module 3: Cognitive Trend Anomaly & Adherence Alert Engine

  1. What exact problem does AI solve? Detects subtle, precipitous drops in cognitive performance or sudden non-adherence patterns that signal medical deterioration, infection (e.g., UTI-induced delirium common in dementia), or medication non-compliance [INFERENCE].
  2. What is the input? 14-day rolling window of daily game scores, task latency distributions, and medication acknowledgment logs.
  3. What is the output? Risk Anomaly Score (0–100) and automated alert flag ("Sudden Attention Score Drop", "3 Missed Hydration Reminders") sent to caregiver [INFERENCE].
  4. Why can a simpler system not solve it? Simple thresholds trigger excessive false alarms due to daily variance. Anomaly detection uses rolling Z-score baselines combined with trend slope evaluation [INFERENCE].
  5. What data is required? Historical patient interaction logs.
  6. How will it be evaluated? False Positive Rate < 10% on simulated test trajectories [INFERENCE].
  7. What happens when it is wrong? Triggers a soft notification to the caregiver ("Check in on grandma today") rather than a panic emergency call [INFERENCE].
  8. Training? Unsupervised baseline statistical modeling (Z-score anomaly / Exponential Weighted Moving Average).
  9. Human Oversight? Mandatory. Caregiver or CHO must review flag before taking clinical action [INFERENCE].

Where AI Should NOT Be Used

  • Critical Reminder Scheduling: Pill schedules, hydration alarms, and doctor appointments MUST use deterministic local system cron schedules (AlarmManager), NOT generative AI or predictive models [INFERENCE].
  • Core Game Rules & Scoring: Correctness of memory matching or pattern recognition MUST be verified by deterministic exact logic, NOT probabilistic ML models [INFERENCE].

14. Data Architecture & Offline/Low-Network Design

Database Schema (Local SQLite / WatermelonDB)

-- Patients Table
CREATE TABLE patients (
    id TEXT PRIMARY KEY,
    name TEXT NOT NULL,
    age INTEGER NOT NULL,
    primary_language TEXT NOT NULL, -- 'as', 'bn', 'mn', 'br', 'en'
    dementia_stage TEXT NOT NULL,  -- 'EARLY', 'MODERATE', 'SEVERE'
    baseline_theta REAL DEFAULT 0.0,
    created_at INTEGER NOT NULL
);

-- GameSessions Table (Local Telemetry)
CREATE TABLE game_sessions (
    id TEXT PRIMARY KEY,
    patient_id TEXT NOT NULL,
    game_type TEXT NOT NULL,        -- 'MEMORY_MATCH', 'ATTENTION_TEA', 'ROUTINE_RECALL', 'PATTERN_WEAVE'
    start_time INTEGER NOT NULL,
    end_time INTEGER NOT NULL,
    score INTEGER NOT NULL,
    max_score INTEGER NOT NULL,
    avg_latency_ms INTEGER NOT NULL,
    hints_used INTEGER DEFAULT 0,
    theta_before REAL NOT NULL,
    theta_after REAL NOT NULL,
    synced_status INTEGER DEFAULT 0, -- 0 = Pending, 1 = Synced
    FOREIGN KEY(patient_id) REFERENCES patients(id)
);

-- CognitiveScores Table (Domain Trends)
CREATE TABLE cognitive_scores (
    id TEXT PRIMARY KEY,
    patient_id TEXT NOT NULL,
    domain TEXT NOT NULL,            -- 'MEMORY', 'ATTENTION', 'EXECUTIVE', 'RECALL'
    current_theta REAL NOT NULL,
    calculated_at INTEGER NOT NULL,
    synced_status INTEGER DEFAULT 0
);

-- Reminders Table
CREATE TABLE reminders (
    id TEXT PRIMARY KEY,
    patient_id TEXT NOT NULL,
    title TEXT NOT NULL,
    category TEXT NOT NULL,          -- 'MEDICINE', 'HYDRATION', 'ACTIVITY', 'APPOINTMENT'
    scheduled_time TEXT NOT NULL,    -- '08:00', '13:00', '20:00'
    audio_prompt_path TEXT,
    is_active INTEGER DEFAULT 1,
    synced_status INTEGER DEFAULT 0
);

-- ReminderLogs Table (Adherence Tracking)
CREATE TABLE reminder_logs (
    id TEXT PRIMARY KEY,
    reminder_id TEXT NOT NULL,
    triggered_at INTEGER NOT NULL,
    acknowledged_at INTEGER,         -- NULL if missed
    status TEXT NOT NULL,            -- 'TAKEN', 'MISSED', 'SNOOZED'
    synced_status INTEGER DEFAULT 0
);

-- SyncQueue Table (Delta Tracking)
CREATE TABLE sync_queue (
    id INTEGER PRIMARY KEY AUTOINCREMENT,
    table_name TEXT NOT NULL,
    record_id TEXT NOT NULL,
    action TEXT NOT NULL,            -- 'CREATE', 'UPDATE', 'DELETE'
    payload_json TEXT NOT NULL,
    created_at INTEGER NOT NULL
);

Delta Synchronization Protocol

CLIENT (WatermelonDB Engine)                        CLOUD BACKEND (NestJS Sync Gateway)
  |                                                      |
  |--- 1. NetInfo detects online connectivity ----------->|
  |                                                      |
  |--- 2. Fetch unsynced records from `sync_queue` ------>|
  |                                                      |
  |--- 3. POST /api/v1/sync/push ----------------------->| (Payload: JSON Delta + last_pulled_at)
  |       Headers: Authorization, Device-HMAC            |
  |                                                      |--- 4. Validate JWT & HMAC
  |                                                      |--- 5. Apply Field Merge & Resolution
  |                                                      |--- 6. Write to PostgreSQL
  |<-- 7. HTTP 200 OK (Server Changes Delta) ------------|
  |                                                      |
  |--- 8. Apply Server Changes to Local SQLite --------->|
  |--- 9. Mark `sync_queue` items as synced (status=1) ->|
  |                                                      |

Conflict Resolution Strategy

  1. Client-Wins for Game Telemetry: The client local device is the absolute authority for game performance, reaction latency, and local theta scores. Server never overwrites client game logs [INFERENCE].
  2. Server/Caregiver-Wins for Schedules: Caregivers modifying medication or doctor appointment schedules on the web portal take precedence over local schedule states [INFERENCE].
  3. Deterministic Timestamp Ordering (Last-Write-Wins): For patient profile edits, the record with the higher ISO-8601 UTC timestamp prevails [RESEARCHED].

15. Security, Privacy & Governance

Compliance & Data Protection Framework

  1. Digital Personal Data Protection (DPDP) Act 2023 Alignment [KNOWN]:

    • Dementia patients are deemed vulnerable individuals. Explicit informed consent must be provided by the nominated primary caregiver (acting as legal guardian) during onboarding [INFERENCE].
    • Data minimization is strictly enforced: No raw microphone audio is recorded or uploaded to cloud servers. Voice STT processing converts speech to text locally or via encrypted streaming, and audio buffers are deleted immediately from memory [INFERENCE].

  2. Ayushman Bharat Digital Mission (ABDM) Integration Readiness [RESEARCHED]:

    • Patient records utilize health ID formatting compatible with ABHA (Ayushman Bharat Health Account) standards [RESEARCHED].
    • All API requests between client app, cloud backend, and ABDM sandbox require TLS 1.3 encryption with SHA-256 HMAC request signing [RESEARCHED].

  3. Data Encryption Standards:

    • At Rest: Local SQLite database encrypted using SQLCipher with AES-256 keys derived from hardware-backed keystore (Android KeyStore System) [RESEARCHED].
    • In Transit: All HTTP communication forced over HTTPS / TLS 1.3 with Certificate Pinning to prevent Man-in-the-Middle (MitM) attacks [RESEARCHED].

  4. Role-Based Access Control (RBAC):

    • Patient Role: Can only access elderly game screens and view local reminder modals; cannot modify caregiver settings or clinical baselines [INFERENCE].
    • Caregiver Role: Authenticated via OTP; can manage schedule, view patient cognitive analytics, receive alerts, and export summary reports [INFERENCE].
    • CHO / ASHA Role: Authenticated via Government Health Portal credentials; can view anonymized or consent-approved multi-patient panels across their designated catchment area [RESEARCHED, INFERENCE].

  5. Medical Disclaimer & Liability Management:

    • The application explicitly displays a non-diagnostic disclaimer: "AASHA-NER is a cognitive stimulation and memory assistance support tool. It is not a diagnostic device and does not replace medical treatment by a certified neurologist." [INFERENCE]

16. MVP Scope (Hackathon Build)

MUST BUILD (Core Hackathon Scope)

  1. Elderly Mobile UI Framework (React Native/Expo):
    • WCAG AAA compliant, high-contrast, large button interface optimized for Android mobile/tablet [INFERENCE].
  2. 3 Core Cognitive Mini-Games with Culturally Familiar NER Motifs:
    • Game 1 (Memory): "Bihu Card Match" (Memory improvement) [PS].
    • Game 2 (Attention): "Tea Leaf Picker" (Selective focus) [PS].
    • Game 3 (Routine Recall): "Daily Life Sequence" (Order daily morning tasks) [PS].
  3. On-Device Adaptive AI Engine:
    • Functional client-side IRT / Elo rating algorithm adjusting game grid size and prompt timing based on user score [PS, RESEARCHED].
  4. Multilingual Voice Assistance (Bhashini Integration):
    • Voice prompt playback and voice navigation supporting Assamese and English (with fallback audio clips) [PS, RESEARCHED].
  5. Local Multi-Modal Reminder Subsystem:
    • Medicine & Hydration alarm pop-ups with local native voice announcements [PS].
  6. Local-First SQLite Database & Offline Sync Engine:
    • WatermelonDB local schema with delta sync queue and connectivity detection [PS, RESEARCHED].
  7. Caregiver Analytics Dashboard (Next.js Web):
    • Real-time dashboard showing cognitive domain radar chart, session log, and missed reminder alerts [PS].

SHOULD BUILD (If Build Time Permits)

  • Game 4 ("Weaving Pattern Match" for pattern/object recognition) [PS].
  • Automated WhatsApp / SMS notification simulation via Twitch/Twilio API when reminders are missed [INFERENCE].
  • Integration of 3rd language (Bengali or Manipuri) into Bhashini voice flow [PS, RESEARCHED].

NICE TO HAVE (Post-Hackathon)

  • Complete 22-language NER dialect coverage (Khasi, Garo, Mizo, Kokborok) [PS, RESEARCHED].
  • Direct integration with Tele-MANAS helpline API (14416) for 1-tap crisis calling [RESEARCHED].
  • Wearable smartwatch integration for automated fall detection and pulse/hydration logging [INFERENCE].

DO NOT BUILD DURING HACKATHON

  • Full ABHA Production Integration (Use mock ABHA Sandbox ID instead) [RESEARCHED].
  • Complex EEG / Brain-Computer Interface hardware connectivity [INFERENCE].
  • Proprietary custom speech AI model training from scratch (Use Bhashini pre-trained APIs) [RESEARCHED].

17. Live Demo Simulation Plan

To deliver a flawless, high-impact demonstration during SIH judging without relying on spotty venue Wi-Fi or unavailable government APIs, the team will implement a transparent, real-time simulation layer:

+---------------------------------------------------------------------------------------------------+
|                                  SIH DEMO SIMULATION FLOW ARCHITECTURE                            |
+---------------------------------------------------------------------------------------------------+

 [SCENARIO ACT 1: PATIENT VOICE EXPERIENCE]
 - Judges see Android Tablet running AASHA-NER in Assamese mode.
 - Student demonstrates hands-free voice launch: "Bihu Khel" (Play Bihu Game).
 - Bhashini STT recognizes Assamese intent -> Launches Memory Game.

 [SCENARIO ACT 2: ON-DEVICE ADAPTIVE AI IN ACTION]
 - Patient deliberately makes 3 mistakes on Level 1.
 - Screen shows live On-Device IRT Log Overlay:
   "Observed Score: 20% | Previous Theta: +0.50 -> Recalculated Theta: -0.25"
 - Game difficulty auto-scales down: Grid size drops from 4x4 to 2x2 in real time (Zero Cloud Delay).

 [SCENARIO ACT 3: OFFLINE ENVIRONMENT & REMINDER TRIGGER]
 - Student toggles Tablet into Airplane Mode (100% Offline).
 - Scheduled 14:00 Hydration Alarm triggers offline: Screen chimes, Assamese audio plays:
   "Pani Khabor Somoy Hoil" (Time to drink water).
 - Student taps "Taken". Event saved to local SQLite `sync_queue` table.

 [SCENARIO ACT 4: ASYNCHRONOUS DELTA SYNC & CAREGIVER ALERT]
 - Student turns off Airplane Mode.
 - Background worker detects network -> Pushes compressed sync payload (4.2 KB) to Cloud Backend.
 - Caregiver Web Dashboard (on judge monitor) updates instantly via WebSocket without page refresh:
   - Cognitive Radar Chart updates.
   - Hydration adherence status changes from "Pending" to "Completed at 14:00".
   - Anomaly Alert Log shows: "Adaptive Engine adjusted difficulty to maintain engagement."

18. Implementation Roadmap

Project Schedule Sized Against Target Deadline: 30 September 2026 [PS]

+---------------------------------------------------------------------------------------------------+
|                                 IMPLEMENTATION ROADMAP (2026)                                     |
+---------------------------------------------------------------------------------------------------+
| PHASE 1: RESEARCH & SPECIFICATION (MAY 15 - JUN 15, 2026)                                         |
| [X] Clinical Psychometric Task Mapping (HMSE/MUDRA)                                               |
| [X] Bhashini API Integration Setup & Audio Asset Recording                                        |
| [X] UI/UX Wireframing for High-Accessibility Elderly Screens                                      |
+---------------------------------------------------------------------------------------------------+
                                                |
                                                v
+---------------------------------------------------------------------------------------------------+
| PHASE 2: CORE MOBILE ENGINE & OFFLINE DB BUILD (JUN 16 - JUL 31, 2026)                             |
| [X] React Native / Expo App Shell Setup with Custom Typography & High Contrast Theme              |
| [X] WatermelonDB Local SQLite Schema Implementation & CRUD Operations                              |
| [X] On-Device 2PL-IRT Adaptive Difficulty Math Library (C++/JS)                                   |
| [X] Native AlarmManager Local Reminder Subsystem Integration                                       |
+---------------------------------------------------------------------------------------------------+
                                                |
                                                v
+---------------------------------------------------------------------------------------------------+
| PHASE 3: GAME MODULES & VOICE INTEGRATION (AUG 01 - AUG 25, 2026)                                |
| [X] Mini-Game 1: Bihu Card Memory Match with Cultural Assets                                      |
| [X] Mini-Game 2: Tea Leaf Selective Attention Focus                                               |
| [X] Mini-Game 3: Daily Routine Sequential Recall                                                  |
| [X] Bhashini Multilingual STT/TTS Integration (Assamese, English)                                 |
+---------------------------------------------------------------------------------------------------+
                                                |
                                                v
+---------------------------------------------------------------------------------------------------+
| PHASE 4: BACKEND SYNC GATEWAY & CAREGIVER PORTAL (AUG 26 - SEP 15, 2026)                          |
| [X] Node.js / NestJS Sync Gateway Development with Protobuf Delta Sync                            |
| [X] PostgreSQL Database Deployment with Prisma ORM                                                |
| [X] Next.js Caregiver Dashboard with Recharts Analytics                                           |
| [X] Automated Anomaly & Adherence Alert Engine                                                    |
+---------------------------------------------------------------------------------------------------+
                                                |
                                                v
+---------------------------------------------------------------------------------------------------+
| PHASE 5: TESTING, SIMULATION & FINAL DEMO HARDENING (SEP 16 - SEP 30, 2026)                       |
| [X] Offline/Online Connectivity Stress Testing (Network Loss during active game)                  |
| [X] System Usability Scale (SUS) Testing with Elderly Volunteers                                  |
| [X] Final SIH Demo Dry Runs & Pitch Deck Finalization                                             |
+---------------------------------------------------------------------------------------------------+

19. Testing & Validation Plan

Test Suites & Evaluation Metrics

  1. Functional Testing:

    • Test Case: Full offline game cycle. Launch app in Airplane Mode, complete 3 game sessions, acknowledge 2 reminders, verify all events persist in local SQLite [INFERENCE].
    • Pass Criteria: 100% data persistence in SQLite; zero app crashes or missing UI state [INFERENCE].

  2. On-Device AI Adaptive Accuracy:

    • Test Case: Simulate 50 consecutive game sessions with synthetic player error profiles (High Ability $\theta=+2.0$, Medium Ability $\theta=0.0$, Low Ability $\theta=-2.0$) [INFERENCE].
    • Pass Criteria: IRT engine converges difficulty selection to keep player success rate between 65%–75% within 5 iterations [RESEARCHED, INFERENCE].

  3. Performance & Latency Testing:

    • Test Case: Measure UI frame rate, local database query latency, and local IRT recalculation execution time on a low-end budget Android phone (2GB RAM) [INFERENCE].
    • Pass Criteria: UI maintains 60 FPS; local DB query < 10ms; IRT calculation < 5ms; RAM footprint < 120 MB [INFERENCE].

  4. Usability Testing (System Usability Scale - SUS):

    • Test Case: Administer app to a test group of 10 elderly users (ages 60+) and measure task completion rate for opening a game and dismissing a pill reminder [RESEARCHED].
    • Pass Criteria: Target SUS score > 75 (Above average usability for senior populations) [RESEARCHED].

  5. False Positive / False Negative Alert Validation:

    • Test Case: Feed 30 days of simulated patient session data containing 3 true cognitive decline events and 10 normal variance days [INFERENCE].
    • Pass Criteria: Zero missed true decline alerts (0% False Negatives); < 1 False Positive alert per month [INFERENCE].

  6. Offline Sync & Resilience Testing:

    • Test Case: Force app kill during active network payload transmission; re-open app and restore connection [INFERENCE].
    • Pass Criteria: Sync queue retries successfully without duplicate record creation on cloud server (Idempotent API design) [RESEARCHED].

20. Risk Register

Risk Event Likelihood Impact Severity Mitigation Strategy Confidence Tag
1. Bhashini API Service Downtime or Network Latency in Field Medium High High Bundle compressed fallback MP3 audio clips for standard UI prompts directly in app bundle; default to Android Native TTS when Bhashini endpoint times out (>2000ms) [RESEARCHED]. [RESEARCHED]
2. Clock Drift on Offline Android Devices High Medium High Use monotonic time intervals (uptime) for session duration logging; compute sync timestamps using server-relative offsets during delta sync reconciliation [RESEARCHED]. [RESEARCHED]
3. Elderly User Frustration / Disengagement Medium High High Enforce local IRT difficulty caps; provide positive multi-sensory feedback (calming folk tunes, visual gold star animations) after every session regardless of score [INFERENCE]. [INFERENCE]
4. Caregiver Device Loss / Unauthorized Access to Health Data Low High Medium Implement local biometric/PIN lock on Caregiver Portal; encrypt SQLite database with SQLCipher AES-256; automatically logout web session after 15 mins of inactivity [RESEARCHED]. [RESEARCHED]
5. Database Corruption During Unexpected Battery Loss Low High Medium Enable SQLite Write-Ahead Logging (WAL mode) in WatermelonDB to guarantee ACID transaction integrity even if device powers off mid-write [RESEARCHED]. [RESEARCHED]

21. Cost & Scalability

Hackathon Phase Costs (Development & Prototype)

  • Mobile Frontend & Local DB: ₹0 (Open-source React Native, Expo, WatermelonDB, SQLite).
  • AI & Voice Services: ₹0 (Bhashini Open API Government Developer Tier) [RESEARCHED].
  • Cloud Hosting: ₹0 (Render / Railway free tier deployment for Node.js + PostgreSQL).
  • Development Tools: ₹0 (VS Code, GitHub, Figma free tier).
  • Total Hackathon Cost: ₹0 (100% Free Open-Source & Sovereign Stack).

Production Rollout Cost Estimate (State-Wide NER Scale - 50,000 Active Patients)

  • Cloud Infrastructure (AWS / NITI Aayog MeghRaj Govt. Cloud):
    • Managed PostgreSQL (High Availability): ~₹12,000 / month [INFERENCE].
    • Node.js Sync Gateway Application Cluster: ~₹18,000 / month [INFERENCE].
    • CDN & Blob Storage for Regional Audio Bundles: ~₹5,000 / month [INFERENCE].
  • Bhashini API Enterprise Gateway: Sponsored under MeitY National Language Translation Mission (NLTM) [RESEARCHED].
  • Hardware & Field Distribution: Leverages existing Android tablets distributed to CHOs and ASHA workers under Ayushman Bharat / National Health Mission (NHM) [RESEARCHED].
  • Estimated Per-Patient Operational Cost: < ₹1.00 per patient / month at scale [INFERENCE].

22. Differentiation

Parameter Generic SIH Submissions AASHA-NER Platform (Proposed Solution)
Architectural Model Standard Cloud-First Web App (Requires active 4G/Wi-Fi). Local-First Reactive Architecture (100% functional offline with WatermelonDB + SQLite) [RESEARCHED].
Difficulty Scaling Basic rule-based level increment (Level 1 -> 2 -> 3). On-Device 2PL-Item Response Theory (IRT) adaptive algorithm running locally without server latency [RESEARCHED].
Language & Voice Layer English/Hindi text-heavy interface with Google Translate. Sovereign Bhashini Voice AI Integration supporting Assamese, Bengali, Manipuri, and Bodo [RESEARCHED].
Cultural Relevance Generic Western puzzles (chess, crosswords, mazes). NER Cultural Motifs (Bihu items, Tea gardens, Muga silk weaving, traditional folk instruments) [PS, RESEARCHED].
Healthcare Integration Isolated standalone student project. Ayushman Arogya Mandir & Tele-MANAS Workflow Alignment with CHO/ASHA monitoring dashboards [RESEARCHED].

23. Future Scope

  1. Expansion of NER Dialect Corpus: Collaborate with North-Eastern Hill University (NEHU) and Gauhati University to expand Bhashini voice corpora for tribal languages like Khasi, Garo, Mizo, and Kokborok [RESEARCHED].
  2. EEG / Wearable Biosensor Integration: Integrate low-cost Bluetooth PPG/EEG headbands to monitor real-time neural engagement and stress levels during gameplay [RESEARCHED].
  3. AI Speech Analysis for Early Dementia Detection: Incorporate acoustic marker analysis (vocal hesitation, pause frequency, phoneme distortion) into voice responses to detect early onset of cognitive decline before clinical symptoms appear [RESEARCHED].
  4. Integration with PM-JAY & State Health Insurance: Link cognitive rehabilitation progress scores to state elderly care support programs across North Eastern states [RESEARCHED].

24. Assumptions & Open Items Log

Consolidated Assumptions [ASSUMPTION]

  1. [ASSUMPTION] Target elderly users have access to an Android smartphone/tablet (Android 8.0+, minimum 2GB RAM) owned by themselves or a co-residing family caregiver.
  2. [ASSUMPTION] Family caregivers will perform initial app setup, select language preferences, and enter medication/hydration schedules.
  3. [ASSUMPTION] Bhashini REST/WebSocket APIs will be accessible during online sync windows with stable response times for text-to-speech generation.
  4. [ASSUMPTION] Synthetic gameplay and patient profile data created for the SIH demo will sufficiently represent real-world clinical dementia progression patterns.

Consolidated Validation Items [NEEDS VALIDATION]

  1. [NEEDS VALIDATION] Exact real-world Word Error Rate (WER) of Bhashini STT models when processing elderly Assamese and Manipuri speech with age-related acoustic variations under rural ambient noise conditions.
  2. [NEEDS VALIDATION] Official process and API access protocol for pushing cognitive health summary reports into the ABDM (Ayushman Bharat Digital Mission) Personal Health Record (PHR) sandbox.
  3. [NEEDS VALIDATION] Field approval requirements from state health authorities (MDoNER / NHM) for distributing cognitive gaming tools via ASHA workers in rural health sub-centres.

Final Recommendation

The team should proceed with building Approach 3: Hybrid Local-First Mobile App with On-Device IRT Adaptive Engine, Bhashini Voice Layer, and Asynchronous Delta Sync. This is the only architecture that strictly honors every explicit requirement in SIH26003 while remaining completely realistic for field deployment in rural North East India [INFERENCE].

Recommended Stack

  • Mobile Frontend: React Native + Expo (SDK 51+) [RESEARCHED]
  • Local Database: WatermelonDB on SQLite engine [RESEARCHED]
  • Adaptive AI Engine: Embedded Client-Side 2PL-IRT / Glicko-2 Library (JS/C++) [RESEARCHED]
  • Voice AI: Digital India Bhashini APIs (Assamese, Bengali, Manipuri, Bodo) [RESEARCHED]
  • Cloud Backend: Node.js (NestJS) + PostgreSQL + Prisma ORM [RESEARCHED]
  • Caregiver Portal: Next.js (v14) + Tailwind CSS + Recharts [INFERENCE]

Biggest Dependency

Availability and API key access for the Digital India Bhashini Division (DIBD) sovereign speech services for North Eastern regional languages [RESEARCHED].

Biggest Risk

Potential Bhashini API service timeouts or voice recognition failures on accented elderly speech, mitigated by bundling local compressed MP3 fallback audio assets [RESEARCHED].

Biggest Technical Challenge

Implementing clean, idempotent bidirectional offline delta synchronization between local WatermelonDB SQLite tables and cloud PostgreSQL without clock-drift corruption or data duplication [RESEARCHED].

First 5 Actions

  1. Action 1: Register for Bhashini API developer credentials on bhashini.gov.in and test Assamese/English STT/TTS REST endpoints [RESEARCHED].
  2. Action 2: Initialize React Native + Expo project with watermelondb and configure local SQLite database schema [RESEARCHED].
  3. Action 3: Implement the on-device 2PL-IRT mathematical function in JavaScript and write unit tests for ability score ($\theta$) updates [RESEARCHED].
  4. Action 4: Build high-accessibility elderly UI design system (64px buttons, high-contrast colors, WCAG AAA compliance) [INFERENCE].
  5. Action 5: Create the 3 core mini-game components with bundled cultural NER visual assets (Bihu cards, Tea leaves, Daily routine items) [PS, RESEARCHED].

Validation Before Coding

  • Validate that watermelondb compiles cleanly on Expo SDK 51 managed workflow with SQLite JSI bindings [RESEARCHED].
  • Verify Bhashini API response latency for Assamese TTS payload generation (< 1500ms) [RESEARCHED].

Buildability Score /100

88 / 100 (Highly buildable by a 4-6 student team using standard JavaScript/TypeScript tooling, open-source databases, and existing sovereign AI APIs within the available timeline) [INFERENCE].

Confidence

HIGH (All underlying technical components—React Native, WatermelonDB SQLite sync, IRT psychometrics, Bhashini AI, NestJS—are well-documented, proven production technologies) [RESEARCHED].

Research Sources

  1. Title: Prevalence of dementia in India: National and state estimates from a nationwide study (LASI-DAD) [RESEARCHED]

    • URL: https://pubmed.ncbi.nlm.nih.gov/36637034/
    • Source Type: Academic Publication / PubMed (2023)
    • Supported Claim: Overall dementia prevalence among adults aged 60+ in India is 7.4% (~8.8 million adults), with higher prevalence in rural areas.
    • Reliability: High (Peer-reviewed national health study).

  2. Title: ICMR Launches MUDRA Toolbox for Dementia Patients in 5 Indian Languages [RESEARCHED]

    • URL: https://www.aninews.in/news/national/general-news/icmr-launches-mudra-toolbox-for-dementia-patients-in-5-indian-languages20211006225846/
    • Source Type: Official News / Indian Council of Medical Research (ICMR)
    • Supported Claim: ICMR MUDRA Toolbox provides culturally and linguistically validated neurocognitive tests for dementia assessment across memory, attention, and executive function.
    • Reliability: High (Official ICMR consortium release).

  3. Title: Digital India Bhashini Division - Voice AI for 22 Indian Languages [RESEARCHED]

    • URL: https://bhashini.gov.in/about-bhashini
    • Source Type: Official Government Portal (MeitY, Govt. of India)
    • Supported Claim: Bhashini provides open APIs for Speech-to-Text, Text-to-Speech, and Translation in 22 scheduled Indian languages including Assamese, Bengali, Manipuri, and Bodo.
    • Reliability: High (Official Government Infrastructure).

  4. Title: Tele-MANAS: National Tele Mental Health Programme [RESEARCHED]

    • URL: https://telemanas.mohfw.gov.in/
    • Source Type: Official Government Portal (MoHFW / NIMHANS)
    • Supported Claim: Tele-MANAS provides 24/7 toll-free mental health support (14416) in 20+ languages across all Indian states including the North East.
    • Reliability: High (Official Government Health Service).

  5. Title: Building Offline-First React Native Apps with WatermelonDB & SQLite [RESEARCHED]

    • URL: https://github.com/nozbe/watermelondb
    • Source Type: Open-Source Software Documentation
    • Supported Claim: WatermelonDB provides reactive, lazy-loaded client-side SQLite database architecture with built-in delta synchronization protocol for offline-first React Native apps.
    • Reliability: High (Widely used production library).

  6. Title: Item Response Theory Analysis of the Clinical Dementia Rating (CDR) [RESEARCHED]

    • URL: https://pmc.ncbi.nlm.nih.gov/articles/PMC8016711/
    • Source Type: Academic Publication / NIH PubMed Central
    • Supported Claim: Item Response Theory (IRT) models latent cognitive ability ($\theta$) and item difficulty parameters ($b_i$) effectively for automated cognitive assessment and adaptive test scaling.
    • Reliability: High (Peer-reviewed psychometric study).
02 / What our

DEVELOPERS SAY

" 30M+ Google Gemini tokens went into researching, analyzing, and understanding all 233 SIH problem statements to make them easier to explore and solve. "
Satyam Kumar
Satyam Kumar
CEO , Full Stack Developer
" The platform makes SIH problem statements much easier to understand. Having everything from the problem details to the procedure in one place. "
Rishabh Ranjan
Rishabh Ranjan
COO , Content Writer
" Oh, you got ads? Well, we need to make money too! But, you can use our platform to help you win SIH and take home the prize money. "
Himanshu Kumar
Himanshu Kumar
CMO , Product Reviewer
03 / Are You

READY TO
BUILD?

Get in touch if you need help bringing your ideas to life, or simply have an interesting idea you'd like to talk through. I'm always open to a conversation and exploring new possibilities together.

LinkedIn
[email protected]