Student Innovation-Ideas that can boost fitness activities and assist in keeping fit.

Description:	Student Innovation-Ideas that can boost fitness activities and assist in keeping fit.
Organization:	AICTE
Department:	AICTE, MIC-Student Innovation
Category:	Hardware
Theme:	Fitness & Sportstd>
Youtube Link:	Not provided
Dataset Link:	Not provided
Contact info:	Not provided

1. Executive Snapshot

This dossier presents a complete technical blueprint, research foundation, and hardware-software system design for Problem Statement SIH26213, issued by AICTE, MIC-Student Innovation under the Hardware category and Fitness & Sports theme [PS]. To address the open-ended mandate of boosting physical activity and assisting citizens in maintaining fitness through student innovation, we propose "FlexiFit IoT"—a modular, ultra-low-cost, smart wearable sensor band coupled with an edge-AI mobile companion application [INFERENCE]. The hardware band utilizes an ESP32-S3 microcontroller with a 6-axis Inertial Measurement Unit (IMU - MPU6050/BNO085) and an optical photoplethysmography (PPG) heart rate sensor (MAX30102) [ASSUMPTION]. Attached to user limbs, dumbbells, or resistance bands, it captures real-time motion, posture, repetition velocity, and cardiovascular exertion [RESEARCHED]. The companion mobile app leverages MediaPipe Pose computer vision for dual-angle posture verification and Edge Impulse TinyML models on-device for real-time rep counting and biomechanical form correction [RESEARCHED]. Tailored for Indian students, gym-goers, and rural youth, the system aligns directly with the Government of India's Fit India Movement and the Ministry of Youth Affairs and Sports' age-appropriate fitness protocols [RESEARCHED].

Metadata Summary

Field	Value
Problem Statement ID	SIH26213 [PS]
Title	Student Innovation-Ideas that can boost fitness activities and assist in keeping fit. [PS]
Organization	AICTE [PS]
Department	AICTE, MIC-Student Innovation [PS]
Category	Hardware [PS]
Theme	Fitness & Sports [PS]
Deadline	30 September 2026 [PS]
Ideas Counter	0/500 [PS]
Dataset Link	None provided (N/A) [PS]

Fundamental System Objective: To build an affordable, open-hardware smart fitness sensor strap with on-device TinyML and computer vision feedback that tracks exercise repetitions, assesses biomechanical posture, prevents workout injuries, and gamifies fitness participation for Indian youth without requiring costly personal trainers or proprietary smart gym equipment [INFERENCE].

2. Problem Deconstruction

The explicit problem statement supplied by AICTE is brief: "Student Innovation-Ideas that can boost fitness activities and assist in keeping fit." [PS]. Because this falls under the Hardware category [PS], an purely software solution will fail judging criteria [INFERENCE]. Below is the atomic deconstruction of explicit requirements, implicit hardware constraints, and technical implications.

Explicit & Derived Requirements

• REQ-01 [Explicit - Category]: Hardware Implementation
  • Source Text: "category": "Hardware" [PS]
  • Interpretation: The core innovation must center around physical hardware (circuit design, sensors, microcontrollers, physical enclosure, or embedded sensing unit) [INFERENCE].
  • Implementation Implication: The team must design, assemble, and demonstrate a physical hardware prototype (e.g., ESP32-S3 microcontroller, IMU sensors, PPG heart-rate sensor, 3D-printed ergonomic casing) alongside the software interface [ASSUMPTION].
  • Confidence Tag: [PS]
• REQ-02 [Explicit - Objective]: Boost Fitness Activities
  • Source Text: "Ideas that can boost fitness activities" [PS]
  • Interpretation: The system must actively motivate, engage, or incentivize users to initiate and maintain daily physical exercise, overcoming inertia, lack of guidance, or boredom [INFERENCE].
  • Implementation Implication: Implement gamification, streaks, local peer leaderboards, guided workout routines based on Fit India protocols, and automated rep logging [INFERENCE].
  • Confidence Tag: [PS]
• REQ-03 [Explicit - Objective]: Assist in Keeping Fit
  • Source Text: "and assist in keeping fit." [PS]
  • Interpretation: The system must provide actionable, real-time safety and biomechanical assistance during exercises (e.g., detecting improper form, preventing spine/knee injuries, measuring intensity) [INFERENCE].
  • Implementation Implication: Integrate real-time posture correction using IMU tilt angle sensing, TinyML motion classification, and camera-based skeletal keypoint analysis (MediaPipe Pose) with instant audio-haptic feedback [RESEARCHED].
  • Confidence Tag: [PS]
• REQ-04 [Inferred - Domain Context]: Alignment with National Fitness Standards
  • Source Text: Derived from AICTE / Ministry of Youth Affairs and Sports context [INFERENCE].
  • Interpretation: The solution should align with India's national health framework, specifically the Fit India Movement's Age-Appropriate Fitness Protocols (18–65 years) [RESEARCHED].
  • Implementation Implication: Incorporate standard baseline tests (push-ups, squat depth test, 1-mile walk/run test, flexibility reach) recommended by the Sports Authority of India (SAI) into the app workflow [RESEARCHED].
  • Confidence Tag: [INFERENCE]

Ambiguities & Missing Information

1. Hardware Scope: The problem statement does not specify whether the hardware should be a wearable, stationary gym attachment, smart mat, or wearable vest [PS]. Resolution: We select a hybrid wearable/equipment-mountable sensor band (wrist/ankle/barbell) as it offers maximum versatility at minimal cost [ASSUMPTION].
2. Target User Group: The statement does not restrict age, setting (home vs. gym), or urban/rural context [PS]. Resolution: Target college students and young adults (18–35 years) practicing home workouts or college gym routines [ASSUMPTION].

3. Stakeholders & User Needs

Stakeholder	Goal	Current Pain	System Need
College Students & Youth	Stay physically fit, build muscle/stamina, and track workout progress without expensive gym memberships or personal trainers [KNOWN].	High cost of personal trainers (₹3,000–₹10,000/month); lack of knowledge on proper posture leading to injury; repetitive workouts become boring [KNOWN].	An affordable (<₹1,500) smart hardware strap that guides workouts, tracks reps, corrects form in real time, and gamifies progress [INFERENCE].
Fitness Enthusiasts / Home Lifters	Execute compound movements (squats, deadlifts, overhead press) safely at home or local gyms [KNOWN].	Inability to check spinal alignment or joint angles during lifting without an observer; injury risk from poor form [RESEARCHED].	Instant haptic/audio alerts during exercise execution when spinal flexion or unsafe joint angles exceed safety thresholds [RESEARCHED].
Physical Education (PE) Instructors / AICTE Colleges	Monitor student physical activity levels and mandate campus-wide fitness routines as per UGC guidelines [RESEARCHED].	Manual attendance tracking; no objective measure of student workout intensity or adherence to Fit India standards [RESEARCHED].	Admin dashboard displaying anonymized aggregate campus activity scores, challenge leaderboards, and fitness baseline trends [INFERENCE].
Sports Authority of India (SAI) / Fit India Program	Promote baseline fitness across Indian citizens using standard age-appropriate fitness protocols [RESEARCHED].	Lack of granular, digital telemetry data on citizen fitness activity and posture safety [RESEARCHED].	Exportable baseline fitness test logs adhering to SAI GOALS protocols (18–65 years) [RESEARCHED].

4. Requirements Traceability

Proposed System Feature	Justification & Requirement Origin	Technical Implementation
ESP32-S3 Wearable Sensor Band (Hardware)	Fulfills REQ-01 [PS] (Hardware Category).	Custom 3D-printed band enclosing ESP32-S3, MPU6050 6-axis IMU, MAX30102 PPG sensor, vibration motor, and LiPo battery [ASSUMPTION].
TinyML On-Device Rep Counter & Motion Classifier	Fulfills REQ-03 [PS] (Assist in keeping fit) & REQ-01 [PS].	Edge Impulse TinyML 1D-CNN classifier compiled into C++ executing directly on ESP32-S3 to recognize exercises and count reps from accelerometer/gyro scope waveforms [RESEARCHED].
MediaPipe Camera Form Analyzer	Fulfills REQ-03 [PS] (Real-time exercise assistance).	Mobile/Web camera feed processed with MediaPipe Pose Landmarker to calculate 3D joint angles (knee, hip, elbow, spine) and flag form errors [RESEARCHED].
Fit India Gamification & Leaderboard Module	Fulfills REQ-02 [PS] (Boost fitness activities) & REQ-04 [INFERENCE].	Mobile UI displaying daily step goals, exercise streaks, peer XP challenges, and Fit India scorecards [RESEARCHED].
Haptic Slouch & Posture Alert Motor	Fulfills REQ-03 [PS] (Injury prevention assistance).	ERM vibration motor triggered on hardware band when IMU pitch angle detects spinal slouching for >3 seconds [RESEARCHED].
Offline Sync & Low-Bandwidth Protocol	Solves rural/low-connectivity deployment constraints [KNOWN].	SQLite local database with Protocol Buffers over BLE, syncing aggregated JSON metrics to cloud backend only when online [INFERENCE].

5. Constraints & Environment Realities

1. Hardware Bill of Materials (BOM) & Cost Constraint: Commercial smart fitness trackers (e.g., Apple Watch, WHOOP) cost between ₹15,000 and ₹35,000, making them inaccessible for most Indian students [KNOWN]. The hardware BOM must stay under ₹1,200–₹1,500 ($15–$18) to be viable for mass adoption [ASSUMPTION].
2. Connectivity & Internet Access: College hostels and rural exercise grounds often suffer from spotty Wi-Fi and unreliable 4G/5G [KNOWN]. The core hardware sensing, rep counting, and posture feedback loop MUST operate 100% offline via BLE and local edge inference [INFERENCE].
3. Power & Battery Realities: Students forget to charge devices daily. The hardware device must utilize ultra-low-power BLE advertising, deep-sleep states, and efficient C++ firmware to sustain at least 8–12 hours of active workout tracking on a 3.7V 500mAh LiPo battery [RESEARCHED].
4. Device Diversity & Mobile Processing: Indian students use a wide spectrum of Android smartphones, ranging from budget Android 10 devices (2GB RAM) to premium models [KNOWN]. Camera-based pose estimation must operate at 15–30 FPS on low-end processors or fallback seamlessly to hardware-only IMU tracking [INFERENCE].
5. Field Conditions & Sweat Resistance: Fitness hardware experiences sweat, impact drops, and rough handling in gyms [KNOWN]. Moisture protection via conformal coating (MG Chemicals 422B) and a TPU flexible 3D-printed case is essential [ASSUMPTION].

6. Research Questions

1. Can a single 6-axis IMU mounted on the wrist or ankle reliably distinguish between different strength exercises (e.g., bicep curls vs. lateral raises vs. squats) without camera aid?
   • Finding: Yes, research demonstrates that 1D-CNN or Random Forest classifiers trained on tri-axial acceleration and gyroscopic angular velocity achieve >88% classification accuracy for repetitive upper and lower body movements [RESEARCHED].
2. How can we prevent double-counting or missed repetitions caused by varying user arm lengths, motion tempos, and partial range-of-motion (ROM)?
   • Finding: Peak-valley detection on bandpass-filtered signal magnitude combined with dynamic thresholding and a minimum rep period window (e.g., 0.8s) eliminates multi-peak noise and tempo variations [RESEARCHED].
3. What are the official fitness evaluation protocols endorsed by the Indian government for youth and adults?
   • Finding: The Ministry of Youth Affairs & Sports published the "Age-Appropriate Fitness Protocols (18-65 Years)" under the Fit India Movement, recommending specific tests: BMI, 60-second Push-ups, Partial Curl-ups, Sit and Reach flexibility, and 2-km Walk/Run [RESEARCHED].
4. What microcontrollers support both low cost (<₹500) and native hardware acceleration for TinyML vector math needed for motion classification?
   • Finding: The ESP32-S3 dual-core Xtensa LX7 microcontroller includes vector instructions (SIMD) designed for neural network matrix multiplication and costs ~₹350–₹450 [KNOWN].

7. Data Requirements & Sources

Data Type	Purpose	Ideal Real-World Source	Hackathon Substitute	Access Method	Availability	Privacy Sensitivity	Confidence
IMU Motion Waveforms	Train rep-counter and exercise classification TinyML model.	UCI Human Activity Recognition Dataset / UCI PAMAP2 [RESEARCHED].	Custom accelerometer + gyro telemetry collected from 5 team members using ESP32 data logger script [ASSUMPTION].	Serial stream / Edge Impulse CLI ingestion [RESEARCHED].	Public / Self-collected [RESEARCHED].	Low (non-identifiable kinetic raw values) [KNOWN].	[RESEARCHED]
Skeletal Pose Landmarks	Compute 3D joint angles for real-time form correction.	MediaPipe 33-Keypoint Pose Model outputs [RESEARCHED].	Real-time camera stream processing via Google MediaPipe Pose JavaScript/Python SDK [RESEARCHED].	Open-source MediaPipe Vision API [RESEARCHED].	Public / Open Source [RESEARCHED].	High (camera video frames must be processed locally in RAM and destroyed) [INFERENCE].	[RESEARCHED]
Heart Rate & PPG Telemetry	Measure workout intensity (Zone 1-5 HR) and calorie expenditure.	Medical Grade ECG / Polar H10 Chest Strap [KNOWN].	MAX30102 PPG sensor using Red/IR LED pulse-oximetry algorithm on ESP32-S3 [RESEARCHED].	I2C bus communication on ESP32 [RESEARCHED].	Open Hardware / Public Arduino drivers [RESEARCHED].	Medium (Biometric health metric) [KNOWN].	[RESEARCHED]
Fit India Benchmark Norms	Score baseline user fitness levels against Indian population benchmarks.	Ministry of Youth Affairs and Sports Fit India GOALS Guidelines [RESEARCHED].	Digitized PDF table lookup integrated into app PostgreSQL/SQLite database [ASSUMPTION].	Manual Extraction from Govt publications [RESEARCHED].	Public Govt Document [RESEARCHED].	None (Public standard) [KNOWN].	[RESEARCHED]

Data Categorization

• A. Explicitly Mentioned Data: None explicitly specified in PS JSON [PS].
• B. Data Required by Inference: 6-axis IMU acceleration ($a_x, a_y, a_z$) and gyroscope ($\omega_x, \omega_y, \omega_z$) streams, PPG photoplethysmography voltage signals, 33-landmark pose coordinates $(x, y, z, \text{visibility})$, user profile age/weight/height [INFERENCE].
• C. Data That Must Be Collected: Training dataset of 500+ exercise repetitions (push-ups, squats, bicep curls, shoulder press) across team members for TinyML model training [ASSUMPTION].
• D. Data Synthetically Generated: Simulated multi-user workout logs, historical telemetry data for load-testing the backend API [ASSUMPTION].
• E. Data Needing Validation: Exact PPG signal accuracy of MAX30102 sensor during heavy motion/sweat artifacts [NEEDS VALIDATION].

8. Existing Solutions & Gap Analysis

Existing Solution	Description	Strengths	Weaknesses	Relevance	Gap Remaining	Source	Confidence
Fit India Mobile App (Sports Authority of India)	Official Govt app for fitness scores, step tracking, and lifestyle goals based on age protocols [RESEARCHED].	Free; official Govt backing; includes age-appropriate fitness protocols and animated videos [RESEARCHED].	Manual exercise logging; no real-time repetition counting; no posture analysis; no hardware integration [RESEARCHED].	High (provides standard protocols to adopt) [RESEARCHED].	Lacks hardware sensing and real-time biomechanical error detection during workouts [INFERENCE].	Google Play Store / PIB [RESEARCHED]	[RESEARCHED]
Commercial Smartbands (e.g., Mi Band, Noise, BoAt)	Consumer wrist-worn fitness trackers providing step counts, HR, and basic activity modes [KNOWN].	Affordable (₹1,500–₹3,000); long battery life; polished industrial design [KNOWN].	Proprietary closed ecosystems; cannot mount on barbells/dumbbells; no rep counting for strength training; no live form correction [RESEARCHED].	Medium (competing consumer form factor) [KNOWN].	Blind to strength training movement quality, joint angles, and barbell path trajectory [INFERENCE].	Commercial specs [KNOWN]	[KNOWN]
AI Computer Vision Apps (e.g., Zenia, Cult.fit AI)	Phone app using front camera to track workouts and count reps [RESEARCHED].	Requires no extra hardware purchase; relies on smartphone camera [RESEARCHED].	Camera view occlusions; fails if user steps out of frame; cannot measure heart rate or kinetic acceleration during heavy lifts [RESEARCHED].	High (software CV comparison) [RESEARCHED].	Camera alone misses kinetic force, velocity, and biometric physiological strain [INFERENCE].	Research Papers / App Stores [RESEARCHED]	[RESEARCHED]

9. Candidate Solution Approaches

+-------------------------------------------------------------------------------------------------+
|                                 CANDIDATE ARCHITECTURE COMPARISON                               |
+-----------------------------------+----------------------------------+--------------------------+
| Approach 1: Hardware-Only         | Approach 2: Mobile Vision Kiosk  | Approach 3: HYBRID EDGE  |
| Sensor Strap (ESP32 + BLE)        | (Phone Camera + Cloud AI)        | (Hardware Strap + Vision)|
| - High portability                | - Zero hardware cost             | - Complete biomechanics  |
| - No visual feedback              | - Occlusion vulnerabilities      | - Dual validation (IMU+CV)|
| - Low cost (₹1,100)               | - High cloud latency/cost        | - Hackathon optimal      |
+-----------------------------------+----------------------------------+--------------------------+

Approach 1: Pure Hardware Wearable (ESP32-S3 + IMU + OLED + Buzzer)

• Mechanism: Standalone wearable band with an ESP32-S3 microcontroller, MPU6050 IMU, OLED display, and vibration motor. Runs local threshold-based rep counting firmware and alerts on slouching via buzzer [RESEARCHED].
• Pros: Highly portable, 100% offline, cheap hardware BOM (~₹1,100) [RESEARCHED].
• Cons: No rich visual UI, limited to simple thresholding, cannot analyze complex multi-joint body postures like deep squat spinal curvature [INFERENCE].
• Hackathon Feasibility: High | Production Feasibility: Medium | Overall Rating: 6.5/10

Approach 2: Pure Vision-Based Mobile App (React Native + MediaPipe + Server AI)

• Mechanism: Mobile smartphone app that uses the front/rear camera feed to analyze pose landmarks using MediaPipe. Video frames sent to cloud API for LLM-based coaching feedback [RESEARCHED].
• Pros: Zero hardware build required; software-only deployment [KNOWN].
• Cons: Fails to meet explicit REQ-01 [PS] (Hardware Category); high camera occlusion issues; heavy battery drain; unusable in tight spaces [INFERENCE].
• Hackathon Feasibility: High (software only) | Production Feasibility: Low (violates problem category) | Overall Rating: 3/10 (Disqualified due to REQ-01 [PS])

Approach 3: Hybrid Hardware-Edge AI System (FlexiFit Hardware Strap + Companion Edge App) [RECOMMENDED]

• Mechanism: Dual-sensing ecosystem. A versatile ESP32-S3 smart hardware sensor band worn on limbs or attached to gym weights captures raw kinetic IMU and PPG heart-rate telemetry via BLE 5.0 [RESEARCHED]. The companion mobile app runs MediaPipe Pose Landmarker for computer vision joint tracking and Edge Impulse TinyML for real-time sensor fusion [RESEARCHED].
• Pros: Satisfies Hardware requirement REQ-01 [PS]; combines kinetic acceleration/HR data with visual camera posture checking; operates completely on edge devices (no cloud latency); highly engaging for SIH judges [INFERENCE].
• Cons: Requires simultaneous hardware embedded programming and mobile app engineering effort [INFERENCE].
• Hackathon Feasibility: High (for 4–6 student team) | Production Feasibility: High | Overall Rating: 9.5/10

10. Recommended Approach & Why

We select Approach 3: Hybrid Hardware-Edge AI System (FlexiFit) [ASSUMPTION].

Why Alternatives Were Rejected

• Approach 1 (Pure Hardware) was rejected because rely strictly on a tiny OLED screen and buzzer provides poor user engagement for gamification and lacks the camera-based body joint angle validation required for detailed form correction [INFERENCE].
• Approach 2 (Pure Vision) was rejected because it directly violates explicit REQ-01 [PS] which mandates a Hardware solution under AICTE MIC Student Innovation [PS]. Furthermore, camera-only tracking cannot measure real-time cardiovascular exertion (heart rate) or micro-vibrations in barbell movement [RESEARCHED].

Winning Synergy

The hybrid approach distributes computational load effectively:

1. Hardware Band (ESP32-S3 Edge): Captures high-frequency (100Hz) kinetic acceleration, angular velocity, and PPG heart rate data. Performs instant haptic vibration alerts for dangerous accelerations or slouching [RESEARCHED].
2. Mobile App (Phone Edge): Receives BLE telemetry and pairs it with optical pose estimation from the phone camera. Displays real-time rep counters, joint angle overlays, form warnings, and Fit India leaderboard scores [RESEARCHED].

11. Technical Architecture

Layered System Architecture Diagram

+---------------------------------------------------------------------------------------------------+
|                                  FLEXIFIT SYSTEM ARCHITECTURE                                     |
+---------------------------------------------------------------------------------------------------+

 [ HARDWARE LAYER - PHYSICAL WEARABLE BAND ]
  +-----------------------------------------------------------------------------------------------+
  | Sensors: MPU6050 (6-Axis IMU) | MAX30102 (PPG Heart Rate) | Flex Sensor                     |
  | Microcontroller: ESP32-S3 (Dual-Core 240MHz, 512KB SRAM, BLE 5.0, Vector Extensions)           |
  | Actuators/UI: ERM Vibration Motor | 0.96" I2C OLED Display | TP4056 USB-C Charger + LiPo     |
  | Firmware: FreeRTOS | ESP-IDF / Arduino C++ | TinyML Edge Impulse C++ Runtime (1D-CNN)         |
  +-----------------------------------------------------------------------------------------------+
                                                 |
                                  BLE 5.0 GATT Telemetry (100Hz IMU / 1Hz HR)
                                                 v
 [ MOBILE / CLIENT LAYER - COMPANION APP ]
  +-----------------------------------------------------------------------------------------------+
  | Framework: React Native / Flutter Cross-Platform Client                                       |
  | Computer Vision Engine: Google MediaPipe Pose Landmarker (33 Skeletal Keypoints @ 30FPS)      |
  | Local AI Engine: TensorFlow Lite / ONNX Runtime (Form Correction Rules + Angle Matrix)        |
  | Local Database: WatermelonDB / SQLite (Offline Cache & Workout Telemetry)                     |
  | Audio Engine: Text-To-Speech (TTS) Voice Coach ("Keep your back straight!", "2 Reps Left")   |
  +-----------------------------------------------------------------------------------------------+
                                                 |
                                     HTTPS REST / WebSockets Sync (JSON)
                                                 v
 [ BACKEND SERVICES LAYER - CLOUD / LOCAL SERVER ]
  +-----------------------------------------------------------------------------------------------+
  | API Gateway: Node.js (Express) / FastAPI Reverse Proxy                                        |
  | Application Core: User Authentication (JWT) | Fit India Challenge Engine | Leaderboards        |
  | Database: PostgreSQL + PostGIS (Spatial Leaderboards) | Redis (Cache & Session)               |
  | Analytics Engine: Python Biomechanics Pipeline (Longitudinal Form Trend Analysis)             |
  +-----------------------------------------------------------------------------------------------+

Data Flow Narrative

1. Motion Acquisition: As the user performs an exercise (e.g., Squat), the hardware band's MPU6050 sensor samples tri-axial acceleration and gyroscopic rates at 100 Hz [RESEARCHED].
2. On-Device Processing & BLE Transmission: The ESP32-S3 applies a Kalman/Mahony filter to calculate smooth Euler pitch/roll angles [RESEARCHED]. If the pitch indicates severe forward slouching (>30° tilt), the ESP32 directly triggers the vibration motor in <20ms [RESEARCHED]. Simultaneously, packaged IMU + PPG packet frames are broadcast via BLE GATT notifications [RESEARCHED].
3. Mobile Sensor Fusion & Vision Analysis: The companion app receives BLE packets while capturing camera frames. MediaPipe Pose extracts keypoint coordinates (e.g., Hip [24], Knee [26], Ankle [28]) [RESEARCHED]. The app calculates the interior knee angle: $$\theta = \arccos\left(\frac{\vec{v}{\text{knee}\to\text{hip}} \cdot \vec{v}{\text{knee}\to\text{ankle}}}{|\vec{v}{\text{knee}\to\text{hip}}| |\vec{v}{\text{knee}\to\text{ankle}}|}\right)$$
4. Rep Counting & Form Validation: When the knee angle drops below 90° and the IMU Z-axis acceleration registers vertical reversal, 1 valid repetition is counted [RESEARCHED]. Audio feedback ("Rep 1 Complete - Good Depth!") is spoken via TTS [INFERENCE].
5. Offline Persistence & Cloud Sync: The workout summary is saved locally in SQLite. When internet connectivity becomes available, the record syncs to PostgreSQL to update the Fit India campus leaderboard [INFERENCE].

12. Technology Stack

Layer	Selected Technology	Why Selected	Alternative Considered	Why Alternative Rejected
Hardware MCU	ESP32-S3-WROOM-1	Integrated BLE 5.0, Wi-Fi, dual-core 240MHz CPU, hardware vector acceleration for TinyML, low cost (~₹400) [KNOWN].	Arduino Uno / Nano	Lacks built-in BLE/Wi-Fi, insufficient SRAM (2KB) for TinyML inference [KNOWN].
Hardware Motion Sensor	MPU6050 / BNO085	High precision 6-axis MEMS accelerometer + gyro, low power, mature open-source libraries [RESEARCHED].	ADXL345	Accelerometer only; lacks gyroscopic orientation data needed for angular velocity tracking [RESEARCHED].
Hardware Heart Rate Sensor	MAX30102	Integrated pulse-oximetry and heart-rate optical sensor with I2C interface [RESEARCHED].	Discrete Photodiode	Requires complex analog filtering circuits, prone to noise during movement [RESEARCHED].
Firmware Framework	ESP-IDF / Arduino C++	Highly optimized, direct hardware register access, native Edge Impulse C++ SDK integration [RESEARCHED].	MicroPython	Higher RAM overhead and slower execution speed for 100Hz DSP filtering [KNOWN].
Mobile App Framework	React Native (v0.74+)	Rapid cross-platform iOS/Android development, rich UI component library, strong BLE native modules [KNOWN].	Native Android (Java/Kotlin)	Slower multi-platform build iteration during 36-hour hackathon window [INFERENCE].
Vision Pose Engine	Google MediaPipe Pose	Ultra-lightweight (runs on-device at 30 FPS), 33 3D skeletal landmarks, no cloud API cost [RESEARCHED].	OpenPose	Extremely heavy GPU requirement; cannot run locally on mobile hardware without severe lag [RESEARCHED].
TinyML Toolchain	Edge Impulse	Automated dataset ingestion, 1D-CNN feature extraction pipeline, direct C++ library export for ESP32-S3 [RESEARCHED].	TensorFlow Lite Micro (manual)	Manual C++ quantization and operator resolution pipeline takes excessive setup time in hackathon [RESEARCHED].
Backend Framework	Node.js (Express) + TypeScript	Asynchronous I/O ideal for lightweight JSON telemetry, fast prototyping [KNOWN].	Python Django	Heavy framework overhead; slower response times for real-time WebSocket messaging [KNOWN].
Database Layer	PostgreSQL + SQLite	SQLite for mobile offline storage; PostgreSQL for cloud multi-user leaderboards and analytics [KNOWN].	MongoDB	Relational schema fits user profiles, structured workout logs, and leaderboard rankings better [INFERENCE].

13. AI/ML Justification

+-------------------------------------------------------------------------------------------------+
|                                 AI / NON-AI DECISION MATRIX                                     |
+------------------------------------+------------------------------------------------------------+
| TASK                               | IMPLEMENTATION STRATEGY                                    |
+------------------------------------+------------------------------------------------------------+
| 1. Repetition Counting             | Rule-Based DSP Peak-Detection (NON-AI) - High precision   |
| 2. Exercise Classification         | TinyML 1D-CNN on ESP32-S3 (AI) - Pattern matching waveforms|
| 3. Skeletal Joint Angle Calculation| Vector Geometry Trigonometry (NON-AI) - Deterministic Math |
| 4. Form Error Detection            | MediaPipe Keypoints + Rule Heuristics (AI + Rule Hybrid)   |
+------------------------------------+------------------------------------------------------------+

Component 1: TinyML Exercise Classification Model (On-Device ESP32-S3)

1. What exact problem does AI solve? Distinguishes between similar limb movements (e.g., differentiating Bicep Curls, Lateral Raises, and Overhead Presses) from raw 6-axis IMU sensor streams [RESEARCHED].
2. What is the input? Time-series array of 6 channels ($a_x, a_y, a_z, \omega_x, \omega_y, \omega_z$) over a sliding window of 2 seconds (200 samples at 100Hz) [RESEARCHED].
3. What is the output? Probability array for class labels: [Bicep_Curl, Lateral_Raise, Overhead_Press, Squat, Idle] [RESEARCHED].
4. Why can a simpler rule/system not solve it? Threshold rules fail when different users execute movements with varying tempos, wrist rotations, or amplitudes [RESEARCHED].
5. What data is required? ~15 minutes of labeled IMU motion data per exercise collected across diverse subjects [RESEARCHED].
6. How will it be evaluated? Categorical cross-entropy loss, Confusion Matrix, and F1-score (>85% target) on test split [RESEARCHED].
7. What happens when it is wrong? App prompts user to manually select exercise type from dropdown UI [INFERENCE].
8. Does it require training? Yes, pre-trained before deployment on Edge Impulse platform [RESEARCHED].
9. Does it require fine-tuning? No on-device fine-tuning needed for hackathon [ASSUMPTION].
10. Does it require RAG? No [INFERENCE].
11. Does it require computer vision? No (this model operates on kinetic IMU data) [RESEARCHED].
12. Does it require speech processing? No [INFERENCE].
13. Does it require forecasting? No [INFERENCE].
14. Is human oversight required? No, user can override classification manually in mobile UI [INFERENCE].

Component 2: MediaPipe Computer Vision Pose Estimation (Mobile Client)

1. What exact problem does AI solve? Extracts 33 3D body keypoint locations $(x, y, z)$ from raw camera frames in real-time [RESEARCHED].
2. What is the input? RGB Camera Frame ($640 \times 480$ image tensor) [RESEARCHED].
3. What is the output? Normalized landmark coordinates for eyes, shoulders, elbows, wrists, hips, knees, ankles [RESEARCHED].
4. Why can a simpler rule/system not solve it? Object tracking without neural keypoint regression cannot handle background clutter, body occlusion, or varying lighting conditions [RESEARCHED].
5. What data is required? Pre-trained Google MediaPipe Pose model weights (bundled in SDK) [RESEARCHED].
6. How will it be evaluated? Mean Percentage of Correct Keypoints ([email protected]) [RESEARCHED].
7. What happens when it is wrong? Low keypoint visibility confidence score ($<0.5$) triggers UI prompt: "Position yourself fully in camera view" [INFERENCE].
8. Does it require training? No (utilizes Google's pre-trained weights) [RESEARCHED].
9. Does it require fine-tuning? No [RESEARCHED].
10. Does it require RAG? No [INFERENCE].
11. Does it require computer vision? Yes, primary CV task [RESEARCHED].
12. Does it require speech processing? No [INFERENCE].
13. Does it require forecasting? No [INFERENCE].
14. Is human oversight required? No [INFERENCE].

Explicit Non-AI Directives

• Repetition Counting: AI/LLM is NOT used for rep counting. Rep counting relies on deterministic signal processing: Butterworth low-pass filtering followed by peak-valley detection on acceleration magnitude $A = \sqrt{a_x^2 + a_y^2 + a_z^2}$ [RESEARCHED].
• Joint Angle Calculation: AI is NOT used to calculate angles. Angles are calculated using standard inverse trigonometric vector dot products on keypoint vectors [RESEARCHED].

14. Data Architecture & Offline/Low-Network Design

Data Schema (SQLite Local / PostgreSQL Backend)

-- User Profile & Health Metrics Table
CREATE TABLE users (
    user_id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    full_name VARCHAR(100) NOT NULL,
    age INT NOT NULL,
    gender VARCHAR(10),
    weight_kg NUMERIC(5,2),
    height_cm NUMERIC(5,2),
    college_institution VARCHAR(150),
    created_at TIMESTAMP WITH TIME ZONE DEFAULT CURRENT_TIMESTAMP
);

-- Fitness Activity Session Log Table
CREATE TABLE workout_sessions (
    session_id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    user_id UUID REFERENCES users(user_id),
    exercise_type VARCHAR(50) NOT NULL, -- 'SQUAT', 'PUSHUP', 'BICEP_CURL'
    start_time TIMESTAMP WITH TIME ZONE NOT NULL,
    end_time TIMESTAMP WITH TIME ZONE,
    total_reps INT DEFAULT 0,
    valid_reps INT DEFAULT 0,
    avg_heart_rate INT,
    form_accuracy_percentage NUMERIC(5,2),
    synced_to_cloud BOOLEAN DEFAULT FALSE
);

-- Detailed Repetition Telemetry Table (Stored Locally)
CREATE TABLE rep_telemetry (
    rep_id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    session_id UUID REFERENCES workout_sessions(session_id),
    rep_number INT NOT NULL,
    duration_seconds NUMERIC(4,2),
    peak_acceleration NUMERIC(6,2),
    min_joint_angle NUMERIC(5,2),
    max_joint_angle NUMERIC(5,2),
    form_flag VARCHAR(50) -- 'GOOD', 'INCOMPLETE_ROM', 'SLOUCH_DETECTED'
);

Offline & Synchronization Strategy

1. Local Persistence First: All raw BLE telemetry packets, rep records, and workout scores are stored in local SQLite storage (WatermelonDB/Room) immediately upon completion [INFERENCE].
2. Bandwidth Reduction via Delta Aggregation: Raw 100Hz IMU stream data is processed in real time and discarded. Only summarized rep metrics (e.g., duration, min/max angles, average HR, error count) are persisted [INFERENCE].
3. Sync Trigger & Retry Mechanism: An automated background sync worker checks for internet availability. When online, pending records (synced_to_cloud = FALSE) are transmitted via compressed HTTP POST JSON payloads with exponential backoff retries [INFERENCE].
4. Conflict Resolution: Client timestamp wins for individual workout sessions. Server recalculates cumulative leaderboard rankings upon ingestion [INFERENCE].

15. Security, Privacy & Governance

1. Personal & Biometric Data Handling: User metrics (age, weight, heart rate) are classified as Sensitive Personal Data [KNOWN]. Data is stored encrypted at rest using AES-256 on local mobile devices [INFERENCE].
2. Camera Feed Privacy Assurance: Camera frames captured during exercise are processed in-memory (RAM) via MediaPipe WASM/Native C++ bindings and immediately discarded [RESEARCHED]. Video frames are NEVER saved to disk, uploaded to cloud servers, or transmitted across network interfaces [INFERENCE].
3. BLE Pair Authorization: Hardware band implements secure BLE pairing using Passkey Entry / Just Works protocol with AES-128 encryption to prevent unauthorized telemetry eavesdropping [RESEARCHED].
4. Regulatory Validation Disclaimer: The device is an educational and fitness lifestyle tracking prototype. It is NOT a certified medical device under CDSCO (Central Drugs Standard Control Organisation) regulations [KNOWN]. The user interface explicitly displays a medical disclaimer: "FlexiFit is a fitness feedback tool and is not intended for diagnostic or medical rehabilitation purposes" [INFERENCE].

16. MVP Scope (Hackathon Build)

+-------------------------------------------------------------------------------------------------+
|                                    MVP SCOPE BOUNDARIES                                         |
+---------------------------------------------------+---------------------------------------------+
| IN SCOPE (MUST BUILD FOR SIH)                     | OUT OF SCOPE (PRODUCTION FUTURE)            |
+---------------------------------------------------+---------------------------------------------+
| - Functional ESP32-S3 + MPU6050 + MAX30102 Band   | - Medical-grade ECG certification           |
| - Live BLE telemetry streaming to Mobile App      | - Custom PCB manufacturing & injection case |
| - MediaPipe 3D Pose estimation for 2 exercises    | - Multi-user cloud synchronization server   |
| - Real-time rep counter & slouch haptic alert     | - iOS App Store publication                 |
| - Local Fit India baseline score dashboard        | - Payment gateway / monetization            |
+---------------------------------------------------+---------------------------------------------+

In Scope

MUST BUILD

• Hardware Sensor Band: ESP32-S3 microcontroller wired with MPU6050 IMU, MAX30102 HR sensor, ERM vibration motor, OLED display, powered by 3.7V battery on breadboard/perfboard casing [ASSUMPTION].
• BLE Firmware: C++ firmware reading IMU at 100Hz, running Mahony orientation filter, broadcasting telemetry over BLE GATT [RESEARCHED].
• Mobile App (Android): React Native / Flutter app receiving BLE data, rendering live camera overlay with MediaPipe skeletal keypoints [RESEARCHED].
• Core Workout Engine: Dual exercise support: Squats (knee angle + vertical displacement) and Bicep Curls (elbow angle + wrist rotation) with live audio rep counter [RESEARCHED].
• Posture Alert: Real-time haptic vibration on hardware band when slouching angle exceeds 30° [RESEARCHED].

SHOULD BUILD

• Fit India Assessment Screen: Digital implementation of SAI GOALS 60-second push-up and sit-and-reach flexibility tests [RESEARCHED].
• Local Leaderboard: SQLite-backed leaderboard showing student performance XP and streak days [INFERENCE].

NICE TO HAVE

• 3D Printed Case Enclosure: Custom CAD casing designed in Fusion360 and printed in PLA/TPU filament [ASSUMPTION].

Described Only / Out of Scope

• Multi-tenant cloud enterprise administrative dashboard [INFERENCE].
• Custom silicon ASIC or mass production PCB fabrication [INFERENCE].
• Support for >10 complex gym exercises [INFERENCE].

17. Live Demo Simulation Plan

To ensure a seamless, foolproof presentation during the SIH judging round without relying on unstable venue Wi-Fi or hardware connectivity glitches, the team will prepare a dual-mode live demo strategy [INFERENCE].

+-------------------------------------------------------------------------------------------------+
|                                     LIVE DEMO EXECUTION PLAN                                    |
+-------------------------------------------------------------------------------------------------+
|  MODE A: PRIMARY LIVE HARDWARE DEMO                                                            |
|  1. Presenter wears physical ESP32-S3 band on arm/leg.                                         |
|  2. Band pairs over Direct BLE to Android Tablet on tripod.                                    |
|  3. Presenter performs Squats in front of Tablet camera.                                        |
|  4. System counts reps aloud, displays live joint angles, and vibrates on intentional slouch. |
+-------------------------------------------------------------------------------------------------+
|  MODE B: HARDWARE SIMULATION FALLBACK (IF BLE / LIGHTING FAILS)                                 |
|  1. Hardware Simulator Script (Python) injects pre-recorded 100Hz IMU CSV telemetry over BLE.  |
|  2. Pre-recorded MP4 workout video fed into MediaPipe camera input pipeline.                    |
|  3. Real-time UI updates identically, demonstrating core math, UI responsiveness, and logic.    |
+-------------------------------------------------------------------------------------------------+

• Simulation Disclosure: The team will explicitly state to judges: "We are now switching to pre-recorded hardware telemetry playback mode to demonstrate our edge algorithm handling edge-case spinal flexions without room interference" [INFERENCE]. Never pretend simulated data is live [PS].

18. Implementation Roadmap

The implementation roadmap is sized against the target deadline of 30 September 2026 [PS].

+-------------------------------------------------------------------------------------------------+
|                                      IMPLEMENTATION ROADMAP                                     |
+-------------------------------------------------------------------------------------------------+
| PHASE 1: HARDWARE PROTOTYPING & FIRMWARE (Weeks 1-3)                                            |
| - Component acquisition (ESP32-S3, MPU6050, MAX30102, LiPo, Vibration motor)                     |
| - Circuit schematic design and breadboard integration                                           |
| - ESP32 C++ firmware: I2C sensor driver setup, 100Hz sampling, BLE GATT service                 |
+-------------------------------------------------------------------------------------------------+
| PHASE 2: TINYML & VISION ENGINE DEVELOPMENT (Weeks 4-6)                                         |
| - Collect 500+ rep IMU dataset; upload to Edge Impulse; train 1D-CNN motion model               |
| - Integrate Google MediaPipe Pose SDK in mobile app; implement joint angle vector math         |
| - Validate haptic slouch trigger latency (<50ms requirement)                                    |
+-------------------------------------------------------------------------------------------------+
| PHASE 3: APP UI & FIT INDIA INTEGRATION (Weeks 7-9)                                             |
| - Develop React Native user interface, dashboard, and live camera canvas overlay                 |
| - Implement SQLite local database, sync manager, and audio TTS feedback engine                  |
| - Build Fit India GOALS protocol testing module and leaderboard screens                        |
+-------------------------------------------------------------------------------------------------+
| PHASE 4: SYSTEM INTEGRATION, CAD CASING & TESTING (Weeks 10-12)                                 |
| - Design 3D printable STL enclosure in Fusion360; print TPU strap case                          |
| - End-to-end system integration testing (Hardware + BLE + Mobile + CV)                           |
| - Execute false-positive/false-negative stress testing and posture boundary validation          |
| - Prepare SIH demo script, simulation fallback toggles, and pitch presentation slides           |
+-------------------------------------------------------------------------------------------------+

19. Testing & Validation Plan

Test Domain	Test Case Objective	Methodology / Test Setup	Success Metric
Functional (Hardware)	Verify 100Hz IMU data sampling stability over BLE.	Connect ESP32 to BLE sniffer (nRF Connect); capture continuous 30-minute telemetry stream [RESEARCHED].	Zero packet loss over 100,000 samples [INFERENCE].
AI / Model Accuracy	Evaluate 1D-CNN exercise classification model.	Test against 100 unseen exercise movement trials across 3 different individuals [RESEARCHED].	$>85%$ classification accuracy on confusion matrix [RESEARCHED].
False-Positive Reps	Ensure non-exercise movements do not trigger rep count.	User drinks water, waves hands, sits down while wearing device [RESEARCHED].	Zero false rep increments recorded [INFERENCE].
False-Negative Reps	Ensure valid deep squats are never missed.	User performs 20 valid squats at varying speeds (slow 3s down vs. fast 1s down) [RESEARCHED].	$\ge 19/20$ ($95%$) valid reps registered [INFERENCE].
Latency Benchmark	Measure delay between posture slouch and haptic vibration alert.	Capture high-speed camera video of user slouching alongside oscilloscope reading motor pin voltage [RESEARCHED].	Total latency $< 100\text{ ms}$ from threshold breach to motor vibration [RESEARCHED].
Offline Performance	Verify system functions without Wi-Fi / cellular data.	Enable Airplane mode on smartphone; execute full 15-minute workout session [INFERENCE].	$100%$ workout data persisted to SQLite local DB without errors [INFERENCE].

20. Risk Register

Risk Event	Likelihood	Impact	Severity	Mitigation Strategy
I2C Bus Lockup on Hardware	Medium	High	High	Implement hardware watchdog timer (WDT) and software I2C bus recovery routine in ESP32 firmware [RESEARCHED].
Bluetooth LE Disconnection During Workout	High	Medium	High	Implement automatic auto-reconnect background thread in mobile app + local RAM buffer on ESP32 to prevent data loss [INFERENCE].
MediaPipe Performance Lag on Low-End Phones	High	High	High	Dynamically downsample camera input frame rate to 15 FPS or fallback to hardware-only IMU rep tracking [INFERENCE].
Optical PPG Noise During Sweaty/Vigorous Motion	High	Medium	Medium	Apply moving average digital bandpass filter on PPG voltage waveform; fallback to static HR measurement between sets [RESEARCHED].
3D Printed Case Mechanical Breakage	Low	Medium	Low	Print enclosure using flexible TPU (Thermoplastic Polyurethane) instead of brittle PLA filament [ASSUMPTION].

21. Cost & Scalability

Hackathon Prototype Cost Breakdown

Component / Item	Unit Quantity	Estimated Cost (INR)
ESP32-S3-WROOM Development Board	1	₹420 [KNOWN]
MPU6050 6-Axis Accelerometer/Gyro Board	1	₹150 [KNOWN]
MAX30102 Pulse Oximeter & Heart Rate Module	1	₹220 [KNOWN]
3.7V 500mAh LiPo Battery + TP4056 Charger	1	₹250 [KNOWN]
0.96" I2C OLED Display + ERM Vibration Motor	1	₹160 [KNOWN]
3D Printing Filament (TPU Enclosure & Band)	1	₹150 [ASSUMPTION]
Total Hackathon Hardware BOM	--	₹1,350 (~$16) [INFERENCE]

Production System Cost & Scalability

• Hardware Scale (10,000 Units): Designing a custom consolidated 4-layer PCB integrating the ESP32-S3 IC, surface-mount IMU, PPG sensor, and power management circuit reduces unit BOM cost to ~₹650–₹750 ($8–$9) [INFERENCE].
• Cloud Infrastructure: Using serverless Node.js containers (AWS Lambda / Google Cloud Run) and managed PostgreSQL (Supabase/Neon), server infrastructure cost per active user is <₹2/month due to offline-first edge architecture [INFERENCE].

22. Differentiation

+-------------------------------------------------------------------------------------------------+
|                                    DIFFERENTIATION MATRIX                                       |
+--------------------------+-----------------------+---------------------+------------------------+
| FEATURE                  | GENERIC HACKATHON APP | COMMERCIAL WATCH    | FLEXIFIT (OUR SOLUTION)|
+--------------------------+-----------------------+---------------------+------------------------+
| Category Compliance      | Software Only (Fails) | Hardware (Closed)   | Open Hardware + App    |
| Biomechanical Form Check | Basic 2D Pose         | None                | Dual IMU + 3D CV       |
| Hardware Cost            | N/A                   | ₹15,000 - ₹35,000   | < ₹1,500 ($18)         |
| Fit India Protocols      | No                    | No                  | Native Built-in        |
| Haptic Slouch Alert      | No                    | Partial (Sedentary) | Real-Time Workout Form |
| Offline Functionality    | Partial               | Yes                 | 100% Offline Edge AI   |
+--------------------------+-----------------------+---------------------+------------------------+

Our solution wins because it specifically targets the Hardware category mandate while bridging physical kinetic sensing with software computer vision [INFERENCE]. Unlike commercial wearables that only log static heart rates, FlexiFit provides active biomechanical injury prevention during heavy compound exercises at a fraction of the cost [INFERENCE].

23. Future Scope

1. Electromyography (EMG) Muscle Strain Integration: Incorporate an analog MyoWare EMG sensor interface into the hardware band to measure active muscle fiber contraction and fatigue thresholds during strength training [RESEARCHED].
2. Multi-Node Mesh Sensing: Expand BLE firmware to support multi-node mesh networks, allowing users to wear 2 bands (e.g., wrist and ankle simultaneously) for full-body kinetic tracking without camera dependence [INFERENCE].
3. Physiotherapy & Rehabilitation Mode: Partner with medical institutions to deploy customized motion tracking protocols for post-stroke or ACL recovery patients [INFERENCE].

24. Assumptions & Open Items Log

Consolidated Assumptions & Needs Validation List

• [ASSUMPTION 01]: Hardware selection of ESP32-S3, MPU6050, MAX30102, ERM motor, and 500mAh LiPo battery fits within the ₹1,500 hackathon budget.
• [ASSUMPTION 02]: A single band worn on the wrist or ankle provides sufficient kinetic acceleration data to count reps for upper and lower body movements respectively.
• [ASSUMPTION 03]: Mobile companion app will be built using React Native targeting Android 10+ devices.
• [ASSUMPTION 04]: 3D printed enclosure will be fabricated using TPU filament for sweat resistance and impact protection.
• [NEEDS VALIDATION 01]: Photoplethysmography (PPG) signal degradation on MAX30102 sensor caused by heavy sweat and motion artifacts during vigorous exercise.
• [NEEDS VALIDATION 02]: Exact MediaPipe 3D keypoint depth coordinate $(z)$ jitter when camera lighting is uneven or back-lit.
• [NEEDS VALIDATION 03]: Battery run-time optimization of ESP32-S3 while maintaining continuous 100Hz IMU sampling and BLE GATT transmission.

Final Recommendation

Build FlexiFit—a dual-node hybrid system combining an ultra-low-cost ESP32-S3 smart hardware sensor band with a MediaPipe-powered mobile edge AI application [INFERENCE]. This directly satisfies the SIH Hardware category, aligns with national Fit India goals, and provides a highly compelling live judging demonstration [INFERENCE].

Recommended Stack

• Hardware: ESP32-S3-WROOM-1, MPU6050 IMU, MAX30102 PPG, ERM Vibration Motor, 0.96" OLED [RESEARCHED].
• Firmware: C++ / ESP-IDF / Arduino IDE, FreeRTOS, Edge Impulse C++ SDK [RESEARCHED].
• Mobile App: React Native, Google MediaPipe Pose, WatermelonDB (SQLite), React Native BLE PLX [RESEARCHED].
• Backend: Node.js, Express, TypeScript, PostgreSQL, Docker [KNOWN].

Biggest Dependency

Stable Bluetooth Low Energy (BLE 5.0) throughput between the ESP32-S3 hardware band and the Android smartphone to maintain low-latency telemetry streaming without packet loss [INFERENCE].

Biggest Risk

Motion artifacts obscuring PPG heart rate readings and camera occlusions causing skeletal keypoint jitter during rapid movements [RESEARCHED].

Biggest Technical Challenge

Real-time sensor fusion synchronizing high-frequency (100Hz) kinetic IMU telemetry with 30 FPS camera keypoint frames in mobile RAM [INFERENCE].

First 5 Actions

1. Procure ESP32-S3 dev board, MPU6050, MAX30102, vibration motor, and LiPo battery components [ASSUMPTION].
2. Flash basic ESP32 C++ firmware to test I2C sensor data acquisition and verify 100Hz streaming over serial monitor [RESEARCHED].
3. Set up Edge Impulse account, log 200 repetitions of Bicep Curls and Squats, and train initial 1D-CNN classifier model [RESEARCHED].
4. Initialize React Native project, integrate MediaPipe Pose Web/Native SDK, and verify live joint angle overlay on camera feed [RESEARCHED].
5. Establish BLE GATT connection between ESP32 hardware and React Native mobile client to display live IMU telemetry [RESEARCHED].

Validation Before Coding

Verify that the MPU6050 IMU pitch/roll angle output remains stable without drift over a 10-minute continuous workout session using the Mahony/Kalman filter [RESEARCHED].

Buildability Score /100

92 / 100 [INFERENCE]. (Reasoning: Components are readily available, open-source libraries exist for ESP32 and MediaPipe, and scope is achievable for a 4–6 student team within the deadline).

Confidence

High (0.90) [INFERENCE].

Research Sources

1. Fit India Movement Official Guidelines

   • Title: Age Appropriate Fitness Protocols and Guidelines (18 to 65 Years)
   • URL: https://yas.gov.in/en/fit-india-fitness-protocols [1.11, 3.1]
   • Source Type: Official Government Portal (Ministry of Youth Affairs and Sports)
   • Supported Claim: Supports national baseline fitness protocol standards and evaluation criteria for Indian citizens [RESEARCHED].
   • Reliability: Highly Reliable (Official Govt Source).

2. Edge AI & TinyML on ESP32 Microcontrollers

   • Title: Edge AI & TinyML on the ESP32: On-Device Machine Learning
   • URL: https://gizantech.com/blog/edge-ai-tinyml-on-esp32-on-device-machine-learning [1.7]
   • Source Type: Technical Engineering Article / Documentation
   • Supported Claim: Validates feasibility of running C++ compiled 1D-CNN Edge Impulse models directly on ESP32-S3 microcontrollers for motion classification [RESEARCHED].
   • Reliability: Reliable.

3. IoT Posture & Exercise Monitoring Research

   • Title: A Portable Smart Fitness Suite for Real-Time Exercise Monitoring and Posture Correction
   • URL: https://www.mdpi.com/1424-8220/21/19/6692 [2.3, 3.6]
   • Source Type: Peer-Reviewed Academic Publication (MDPI Sensors Journal)
   • Supported Claim: Confirms >88% exercise form accuracy using wearable IMU sensors paired with mobile real-time posture feedback [RESEARCHED].
   • Reliability: Highly Reliable (Peer-reviewed academic research).

4. Google MediaPipe Vision Solutions

   • Title: Pose Landmark Detection Guide - Google AI Edge
   • URL: https://developers.google.com/edge/mediapipe/solutions/vision/pose_landmarker [4.8]
   • Source Type: Official Technology Documentation (Google Developers)
   • Supported Claim: Provides technical specification for extracting 33 3D skeletal keypoint landmarks on mobile edge devices in real time [RESEARCHED].
   • Reliability: Highly Reliable (Official Vendor Documentation).

5. Fit India Mobile App Details

   • Title: Fit India Mobile App - Apps on Google Play
   • URL: https://play.google.com/store/apps/details?id=com.sai.fitIndia [4.1]
   • Source Type: Official App Portal (Sports Authority of India)
   • Supported Claim: Documents existing state-of-the-art features and identifies functional gaps regarding real-time hardware exercise tracking [RESEARCHED].
   • Reliability: Highly Reliable.