Integrated Polar Science Outreach, Knowledge Repository and Media Dissemination Portal

1. Executive Snapshot

Executive Summary

The National Centre for Polar and Ocean Research (NCPOR) [PS], an autonomous institute under the Ministry of Earth Sciences (MoES) [PS], conducts cutting-edge research across Earth's polar and oceanic frontiers, including Antarctica (Maitri, Bharati stations) [RESEARCHED], the Arctic (Himadri station) [RESEARCHED], the Himalayas [RESEARCHED], and the Southern Ocean [RESEARCHED]. Despite generating vast troves of scientific datasets, expedition reports, research papers, and multimedia, these assets remain fragmented across legacy systems like the National Polar Data Center (NPDC) [RESEARCHED] and physical library archives [RESEARCHED].

To bridge this gap, PolarIS (Polar Information & Synergy Portal) is proposed—an integrated, web-native platform that functions simultaneously as a FAIR-compliant (Findable, Accessible, Interoperable, Reusable) digital knowledge repository [INFERENCE], an automated media dissemination engine [PS], and an immersive public outreach portal [PS]. PolarIS leverages a decoupled microservices architecture [INFERENCE] featuring an automated AI Content Studio [INFERENCE] that converts dense scientific reports into multi-channel public communications (press releases, infographics text, social media snippets for X/Twitter, LinkedIn, and Instagram) [INFERENCE], alongside an offline-first mobile/field sync client designed specifically for bandwidth-constrained satellite environments at polar bases [INFERENCE].

Metadata Table

Fundamental System Objective

To deliver a unified, open-access polar knowledge platform that automates multi-format scientific content dissemination across public and digital media while providing a robust, FAIR-compliant repository for India's polar research assets [INFERENCE].

2. Problem Deconstruction

The problem statement description explicitly mandates: "Develop a comprehensive outreach portal that archives expedition reports, scientific datasets, publications, photographs, videos and institutional activities while generating content for websites and social media." [PS]

This explicit mandate is deconstructed into seven core atomic requirements:

• Req 1: Expedition Report Archival

  • Source Text: "...archives expedition reports..." [PS]
  • Interpretation: Provide a structured digital asset store for past and present Indian Scientific Expeditions to Antarctica (ISEA), Arctic expeditions, Himalayan cryosphere studies, and Southern Ocean cruises [INFERENCE].
  • Implementation Implication: Full-text indexing, PDF previewer, metadata cataloging (expedition year, voyage leader, leader report, station, discipline) [INFERENCE].
  • Confidence Tag: [PS]

• Req 2: Scientific Dataset Archival & Discovery

  • Source Text: "...archives... scientific datasets..." [PS]
  • Interpretation: Catalog raw and processed research datasets (meteorological, oceanographic, glaciological, biological, paleoclimate) with spatial and temporal bounds [INFERENCE].
  • Implementation Implication: Implementation of international polar metadata standards (ISO 19115 / Schema.org Dataset JSON-LD) [RESEARCHED] with download links, dataset landing pages, and spatial bounding box search [INFERENCE].
  • Confidence Tag: [PS]

• Req 3: Scientific Publication Archival

  • Source Text: "...archives... publications..." [PS]
  • Interpretation: Store, index, and render peer-reviewed journal papers, technical publications, book chapters, and research bulletins originating from NCPOR scientists [INFERENCE].
  • Implementation Implication: Automated DOI lookup via Crossref API [KNOWN], citation generation, PDF parsing, author attribution, and research theme tagging [INFERENCE].
  • Confidence Tag: [PS]

• Req 4: Multimedia Asset Archival (Photographs & Videos)

  • Source Text: "...archives... photographs, videos..." [PS]
  • Interpretation: High-capacity Digital Asset Management (DAM) system for high-resolution field photography and video footage from research vessels and polar stations [INFERENCE].
  • Implementation Implication: Automatic EXIF/IPTC metadata extraction [KNOWN], video transcoding for adaptive streaming [KNOWN], thumbnail generation, auto-tagging using Computer Vision [INFERENCE], and copyright/licensing watermarking [INFERENCE].
  • Confidence Tag: [PS]

• Req 5: Institutional Activity & Event Archival

  • Source Text: "...archives... institutional activities..." [PS]
  • Interpretation: Repository for institutional events such as World Day for Glaciers, Antarctic Treaty Consultative Meetings (ATCM), student workshops, webinars, and official MoES releases [RESEARCHED].
  • Implementation Implication: Event calendar, photo galleries, press release archive, and event summary logging [INFERENCE].
  • Confidence Tag: [PS]

• Req 6: Automated Web & Social Media Content Generation

  • Source Text: "...while generating content for websites and social media." [PS]
  • Interpretation: Automated/assisted content synthesis engine that ingests complex research papers or expedition logs and outputs formatted press releases, web news articles, short social media captions, thread posts, and relevant hashtags [INFERENCE].
  • Implementation Implication: Fine-tuned Multimodal LLM pipeline [INFERENCE] with human-in-the-loop validation, exportable to JSON/REST formats for X/Twitter, LinkedIn, Facebook, and CMS RSS feeds [INFERENCE].
  • Confidence Tag: [PS]

• Req 7: Public Smart Education & Outreach Portal

  • Source Text: Categorized under "Smart Education" theme; "comprehensive outreach portal" [PS]
  • Interpretation: Public-facing interactive web portal designed for students, educators, and citizens to explore polar science interactively [INFERENCE].
  • Implementation Implication: Interactive 3D/GIS polar maps (Bharati, Maitri, Himadri), virtual station tours, educational quizzes, science blogs, and downloadable educational collateral [INFERENCE].
  • Confidence Tag: [PS] / [INFERENCE]

3. Stakeholders & User Needs

4. Requirements Traceability

5. Constraints & Environment Realities

• Geographic & Operating Environment: NCPOR manages operations across extreme remote geographies: Antarctica (Bharati station at 69°S in Larsemann Hills; Maitri station at 70°S in Schirmacher Oasis) [RESEARCHED], Arctic (Himadri station at 78°N in Ny-Ålesund, Svalbard) [RESEARCHED], Himalayas (Sutri Dhaka base) [RESEARCHED], and oceanographic vessels (ORV Sagar Kanya, Sagar Nidhi) [RESEARCHED].
• Satellite Connectivity Realities: Maitri station relies on restricted SATCOM bandwidth (~4 MHz satellite link) [RESEARCHED]. Bharati has dedicated C-band satellite links connected via NRSC Hyderabad [RESEARCHED], but high latency (>600 ms) and severe blizzard-induced outages remain common [INFERENCE]. Uploading 1 GB video files directly over SATCOM is infeasible [INFERENCE].
• Field Device Realities: Expedition scientists use ruggedized laptops and personal cameras under extreme ambient temperatures (-40°C to +5°C) [RESEARCHED]. Storage drives are filled locally during field traverses [INFERENCE].
• Language & Accessibility: Public outreach in India requires bilingual support (Hindi and English) [KNOWN] to comply with Union official language guidelines and achieve nationwide reach across rural and urban schools [INFERENCE]. WCAG 2.1 AAA accessibility (high contrast, screen reader compatibility) is mandatory for government portals [KNOWN].

6. Research Questions

1. Data Interoperability: How can the metadata schema align with both national standards (NPDC) and global federated polar networks (POLDER, SCAR Antarctic Metadata Directory, NSIDC) without requiring duplicate data entry? [RESEARCHED]
2. Low-Bandwidth Transfer: What client-side compression algorithms (AVIF for images, H.265/AV1 for video) and protocol choices (resumable chunked uploads over HTTP/3 or WebRTC) minimize bandwidth consumption over polar SATCOM links? [INFERENCE]
3. AI Quality & Accuracy: How can we prevent hallucinations when an LLM auto-generates press releases or social media captions from highly technical polar science papers? [INFERENCE]
4. Copyright & Licensing: What content governance mechanisms are needed to handle sensitive pre-publication research data versus public domain educational assets? [INFERENCE]

7. Data Requirements & Sources

8. Existing Solutions & Gap Analysis

9. Candidate Solution Approaches

Approach 1: Monolithic Web CMS with Manual Social Media Publishing

• Mechanism: Standard WordPress / Drupal platform extended with custom plugin modules for document upload and social media embeds.
• Pros: Low initial development complexity; quick template setup during hackathon.
• Cons: Extremely poor handling of large scientific datasets, no native spatial search (GIS), high security vulnerability profile, no automated AI content generation, unsuited for low-bandwidth satellite sync.
• Feasibility Score: Hackathon: 80/100 | Production: 30/100.

Approach 2: Microservices FAIR Repository + Manual REST API Dissemination

• Mechanism: React frontend connected to Django/FastAPI microservices, PostgreSQL with PostGIS for spatial metadata, and Elasticsearch for document search. Social media posts created manually by admins.
• Pros: Clean architecture, high stability, excellent FAIR data standard compliance.
• Cons: Misses the central problem statement requirement for automated social media/website content generation; high manual workload on outreach staff remains.
• Feasibility Score: Hackathon: 70/100 | Production: 65/100.

Approach 3: Decoupled FAIR Knowledge Repository + Multimodal AI Content Studio + Low-Bandwidth Field Sync (RECOMMENDED)

• Mechanism: Next.js single-page frontend (pwa-enabled), FastAPI backend, PostgreSQL + pgvector + PostGIS database, MinIO object storage, FFmpeg video processing pipeline, and an LLM-powered Multimodal AI Studio that ingests papers/media to auto-generate multi-platform content (press releases, X threads, LinkedIn posts, website news snippets). Includes an offline-first service worker for field scientists.
• Pros: Directly addresses all explicit requirements (archival + content generation + outreach) [PS], optimized for low-bandwidth SATCOM, highly scalable, state-of-the-art UI/UX.
• Cons: Higher architectural complexity requiring strict AI guardrails.
• Feasibility Score: Hackathon: 92/100 | Production: 95/100.

10. Recommended Approach & Why

Approach 3 (Decoupled FAIR Knowledge Repository + Multimodal AI Content Studio) is chosen as the winning architecture.

Why Alternatives Were Rejected:

• Approach 1 (Monolithic CMS) was rejected because it fails to support scientific dataset schemas (ISO 19115), lacks spatial GIS query capabilities, and cannot handle offline SATCOM synchronization required at Maitri/Bharati [INFERENCE].
• Approach 2 (Traditional Microservices without AI) was rejected because it ignores the explicit mandate: "while generating content for websites and social media" [PS]. Forcing NCPOR outreach staff to manually write posts for hundreds of papers creates an unscalable bottleneck [INFERENCE].

Approach 3 uniquely balances robust scientific archival standards with modern AI-driven public engagement [INFERENCE].

11. Technical Architecture

Layered Architecture Diagram

+-----------------------------------------------------------------------------------+
|                            PUBLIC OUTREACH & DISSEMINATION                        |
|  +-----------------------+  +--------------------------+  +--------------------+  |
|  | Smart Education Hub   |  | Interactive Polar Map    |  | Media Press Room   |  |
|  | (Quizzes/Virtual Base)|  | (Bharati/Maitri/Himadri) |  | (RSS / Social PR)  |  |
|  +-----------------------+  +--------------------------+  +--------------------+  |
+-----------------------------------------------------------------------------------+
                                         |  (HTTPS / REST / GraphQL)
+-----------------------------------------------------------------------------------+
|                            APPLICATION LAYER (Next.js 14 / React)                 |
|  +-----------------------+  +--------------------------+  +--------------------+  |
|  | Knowledge Repository  |  | AI Content Studio        |  | Field Upload Sync  |  |
|  | (Search/Filter UI)    |  | (Human-in-Loop Review)   |  | (PWA / Offline)    |  |
|  +-----------------------+  +--------------------------+  +--------------------+  |
+-----------------------------------------------------------------------------------+
                                         |  (JWT Auth / REST APIs)
+-----------------------------------------------------------------------------------+
|                            BACKEND SERVICES (FastAPI / Python 3.12)               |
|  +--------------------+ +--------------------+ +----------------+ +---------------+ |
|  | Document & Data    | | Media Processing   | | Spatial GIS    | | AI Content    | |
|  | Archival Service   | | Engine (FFmpeg)    | | Query Engine | | Synthesis    | |
|  +--------------------+ +--------------------+ +----------------+ +---------------+ |
+-----------------------------------------------------------------------------------+
                                         |
+-----------------------------------------------------------------------------------+
|                            DATA STORAGE & INDEXING LAYER                          |
|  +---------------------------------+  +----------------------+  +-----------------+ |
|  | PostgreSQL + PostGIS + pgvector |  | MinIO Object Storage |  | Redis Cache     | |
|  | (Metadata, Vector Embeddings)   |  | (Raw Datasets/Videos)|  | (Job Queue/Task)| |
|  +---------------------------------+  +----------------------+  +-----------------+ |
+-----------------------------------------------------------------------------------+
                                         |
+-----------------------------------------------------------------------------------+
|                            EXTERNAL INTEGRATION & INTEROPERABILITY                |
|  +-----------------------+  +--------------------------+  +--------------------+  |
|  | Schema.org / JSON-LD  |  | Crossref DOI API         |  | Social Media APIs  |  |
|  | (POLDER/SCAR Harvest) |  | (Metadata Sync)          |  | (X, LinkedIn, FB)  |  |
|  +-----------------------+  +--------------------------+  +--------------------+  |
+-----------------------------------------------------------------------------------+

Data-Flow Narrative

1. Ingestion & Archival Path:

   • An expeditioner at Bharati or an administrator in Goa uploads a document (Expedition Report PDF, Research Paper, or Dataset CSV) via the web or offline PWA client.
   • The file is sent to FastAPI backend; media files (photos/videos) pass through FFmpeg for thumbnail extraction, web-optimization (WebP/MP4), and EXIF spatial/temporal parsing [INFERENCE].
   • Text documents are processed via PyMuPDF/Tesseract OCR [KNOWN] to extract raw text, which is chunked and stored in PostgreSQL alongside vector embeddings generated via sentence-transformers in pgvector [INFERENCE].
   • Dataset metadata is assigned an ISO 19115 / Schema.org JSON-LD tag and indexed in PostGIS with spatial coordinates [INFERENCE].

2. AI Dissemination Path:

   • When a new scientific publication or report is archived, the AI Content Studio triggers a job queue task in Redis/Celery [INFERENCE].
   • An LLM processes the abstract and executive summary using a strict prompt template to generate:
     • A 3-tweet summary thread with relevant hashtags (#Antarctica #ClimateChange #MoES) [INFERENCE].
     • A professional LinkedIn press post highlighting key findings [INFERENCE].
     • A concise bilingual web news release (English & Hindi) [INFERENCE].
   • The generated draft enters the Human-in-the-Loop Review Dashboard, where an NCPOR Media Officer approves or edits the content before one-click publishing via social media APIs or RSS [INFERENCE].

12. Technology Stack

13. AI/ML Justification

1. AI Component: Automated Media & PR Synthesizer

• Purpose: Summarize complex scientific papers/expedition reports into accessible public communications (X threads, LinkedIn posts, bilingual press releases) [PS].
• Model Type: Fine-tuned Multimodal LLM (e.g., Llama-3-70B-Instruct or GPT-4o-mini API) [INFERENCE].
• Input: Research paper abstract, PDF text chunks, key figure captions [INFERENCE].
• Output: Structured JSON containing: tweet_thread, linkedin_post, press_release_en, press_release_hi, suggested_hashtags [INFERENCE].
• Data Required: 200+ historical MoES press releases and polar science abstracts for fine-tuning/few-shot prompts [INFERENCE].
• Why AI is Preferable: Writing 5 distinct media formats for dozens of monthly papers manually takes outreach staff ~4 hours per paper. AI completes initial drafts in 5 seconds [INFERENCE].
• Evaluation Metric: ROUGE score against human PR summaries; factual precision check [INFERENCE].
• Failure Mode: Hallucinated scientific figures or misattributed findings.
• Human Oversight: Mandatory. Posts CANNOT auto-publish; they remain in "Pending Review" queue until approved by an NCPOR admin [INFERENCE].

2. AI Component: Auto-Tagging & Visual Search for Polar Media

• Purpose: Automatically categorize uploaded expedition photos and videos (e.g., "Bharati Station", "Penguins", "Glacier Calving", "ORV Sagar Kanya") [INFERENCE].
• Model Type: CLIP (Contrastive Language-Image Pre-training) / Vision Transformer [INFERENCE].
• Input: High-resolution field photos/video frames [INFERENCE].
• Output: Auto-generated descriptive tags and vector embeddings for semantic visual search ("Find photos of ice coring at Maitri") [INFERENCE].
• Data Required: Pre-trained CLIP model zero-shot inference [INFERENCE].
• Why AI is Preferable: Scientists upload thousands of field photos without adding manual tags [INFERENCE].
• Failure Mode: Misidentifying glacier types or animal species.
• Human Oversight: Admins can edit or remove tags in the DAM portal.

Explicit Non-AI Rule

AI MUST NOT be used for:

1. Direct alteration of scientific dataset numerical values or raw expedition log records [INFERENCE].
2. Determining dataset licensing or access permissions [INFERENCE].
3. Generating authoritative scientific conclusions without peer-reviewed source backing [INFERENCE].

14. Data Architecture & Offline/Low-Network Design

Core Database Schema (Entity Relationship Outline)

-- Core Table: Digital Assets (Reports, Papers, Media, Datasets)
CREATE TABLE assets (
    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    title VARCHAR(500) NOT NULL,
    asset_type VARCHAR(50) CHECK (asset_type IN ('EXPEDITION_REPORT', 'DATASET', 'PUBLICATION', 'PHOTO', 'VIDEO', 'EVENT')),
    description TEXT,
    publication_date DATE,
    expedition_name VARCHAR(100), -- e.g., '43rd ISEA', 'Arctic 2023'
    location_name VARCHAR(100),   -- e.g., 'Maitri', 'Bharati', 'Himadri'
    spatial_coverage GEOMETRY(Polygon, 4326), -- PostGIS Spatial Bounding Box
    author_details JSONB,         -- Array of authors, affiliations, ORCIDs
    doi VARCHAR(100),
    file_path VARCHAR(1000) NOT NULL,
    file_size_bytes BIGINT,
    metadata_jsonld JSONB,       -- Schema.org / ISO 19115 compliant JSON
    is_public BOOLEAN DEFAULT TRUE,
    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
);

-- AI Generated Social Media Content Queue
CREATE TABLE ai_generated_content (
    id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    asset_id UUID REFERENCES assets(id) ON DELETE CASCADE,
    platform VARCHAR(50) CHECK (platform IN ('X_TWITTER', 'LINKEDIN', 'WEBSITE_NEWS', 'PRESS_RELEASE')),
    language VARCHAR(10) DEFAULT 'en',
    generated_text TEXT NOT NULL,
    status VARCHAR(30) DEFAULT 'PENDING_APPROVAL', -- 'PENDING_APPROVAL', 'APPROVED', 'PUBLISHED', 'REJECTED'
    reviewer_comments TEXT,
    created_at TIMESTAMP DEFAULT CURRENT_TIMESTAMP
);

Low-Bandwidth & Offline Field Synchronization Architecture

To solve the constraint of ~4 MHz SATCOM satellite links at Maitri base [RESEARCHED]:

1. Client-Side Heavy Compression: Before upload, the PWA web app uses WebAssembly (Wasm) [KNOWN] modules to compress images to AVIF format (70% smaller than JPEG) and transcode high-res video down to 720p H.265/AV1 streams locally on the scientist's laptop [INFERENCE].
2. Resumable Chunked Uploads: Files are split into 1 MB chunks using tus.io protocol [KNOWN]. If a blizzard disrupts the satellite link, the upload pauses seamlessly and resumes from the exact byte offset when connectivity returns [INFERENCE].
3. Local IndexedDB Caching: Expedition logs written offline are stored in browser IndexedDB [KNOWN]. A Background Sync Service Worker detects network availability and pushes batched metadata JSON payload first (few kilobytes) while deferring binary file transfers to off-peak satellite windows [INFERENCE].

15. Security, Privacy & Governance

• Personal Data & Privacy: Scientist profiles contain names, institutional affiliations, and ORCIDs [RESEARCHED]. No sensitive personal contact information or health/medical records from expeditioners will be publicly exposed [INFERENCE].
• Data Access & Role-Based Access Control (RBAC):
  • Role 1: Public User / Student - Read-only access to public reports, published datasets, media, news, and outreach tools [INFERENCE].
  • Role 2: Contributor / Scientist - Upload assets, edit own dataset metadata, queue field uploads [INFERENCE].
  • Role 3: Media Officer / Admin - Approve/edit AI-generated social media posts, manage user permissions, publish content to live channels [INFERENCE].
• Data Governance & Embargo Support: Research datasets subject to pre-publication embargo periods (e.g., 12 months) are locked under restricted access, exposing only discovery metadata until the embargo release date [INFERENCE].
• Compliance: Fully compliant with the Indian Antarctic Act, 2022 [RESEARCHED], MoES Data Sharing Policy [RESEARCHED], and Union Government Cybersecurity Guidelines (CERT-In) [KNOWN].
• Encryption: All data in transit enforced via TLS 1.3; data at rest in MinIO and PostgreSQL encrypted using AES-256 [KNOWN].

16. MVP Scope (Hackathon Build)

Given the build deadline of 30 September 2026 [PS], the team must strictly prioritize features that prove core value:

In Scope (MUST BUILD)

1. Unified Digital Knowledge Repository UI: Search and filter engine across Expedition Reports, Papers, Datasets, and Media Assets with instant PDF viewer [INFERENCE].
2. AI Dissemination Studio: Working LLM pipeline taking a paper/report abstract as input and generating editable X/Twitter threads, LinkedIn posts, and press release drafts [INFERENCE].
3. Spatial Dataset Map Search: Leaflet-based interactive map displaying dataset markers and station coordinates (Bharati, Maitri, Himadri) [INFERENCE].
4. Smart Education Outreach Hub: Interactive "Polar Explorer" page featuring virtual base information, interactive ice-core timeline, and educational quizzes [INFERENCE].
5. Schema.org Metadata Export: Automatic JSON-LD metadata generation for dataset landing pages to prove FAIR compliance [INFERENCE].

Described Only / Out of Scope (HACKATHON EXCLUSIONS)

• Direct live auto-posting integration with real NCPOR social media API credentials (simulated via manual download/copy or sandbox webhooks during demo) [INFERENCE].
• Full historical migration of all thousands of physical legacy physical reports (demonstrated using a representative sample of ~50 processed PDFs) [INFERENCE].
• Hardware SATCOM satellite physical testing (simulated using Chrome DevTools network throttling) [INFERENCE].

17. Live Demo Simulation Plan

To present a flawless, realistic demonstration without requiring live government credentials or polar satellite hardware:

1. Simulating Polar SATCOM Uploads:
   • Method: Use Chrome DevTools network emulation set to "Custom Low-Bandwidth Profile" (500 Kbps bandwidth, 800ms latency) [INFERENCE].
   • Demo: Show a scientist uploading a 50MB expedition media file at Maitri station. Demonstrate client-side Wasm image compression reducing size to 3MB, chunked upload over tus.io, simulated disconnect, and automatic resumption [INFERENCE].
2. Simulating AI Content Generation Live:
   • Method: Upload a real, highly technical NCPOR research PDF (e.g., "Aerosol optical depth measurements at Maitri") [RESEARCHED].
   • Demo: Click "Generate Outreach Package". Live-stream the AI output generating a 3-tweet thread with hashtags, a LinkedIn post, and a Hindi press release. Show the admin editing a word in the review panel and approving it for distribution [INFERENCE].
3. Simulating FAIR Metadata Harvesting:
   • Method: Navigate to a dataset landing page and run Google's Rich Results Test tool live to validate the embedded Schema.org/Dataset JSON-LD structure [INFERENCE].

18. Implementation Roadmap

The roadmap is structured leading up to the 30 September 2026 deadline [PS]:

+-----------------------------------------------------------------------------------+
| PHASE 1: FOUNDATION & SCHEMA DESIGN (Months 1 - 2)                                |
| - Finalize PostgreSQL/PostGIS database schema & MinIO bucket structures.          |
| - Scrape & convert sample NCPOR reports, publications, and open datasets.        |
+-----------------------------------------------------------------------------------+
                                         |
+-----------------------------------------------------------------------------------+
| PHASE 2: CORE REPOSITORY & MEDIA PIPELINE (Months 3 - 4)                          |
| - Build FastAPI backend endpoints & Next.js search/filter frontend.               |
| - Integrate FFmpeg media processing and PDF full-text indexing with pgvector.     |
+-----------------------------------------------------------------------------------+
                                         |
+-----------------------------------------------------------------------------------+
| PHASE 3: AI DISSEMINATION STUDIO & OUTREACH PORTAL (Months 5 - 6)                 |
| - Build LangChain/LLM content generation pipeline with strict JSON guardrails.    |
| - Implement Human-in-the-Loop review dashboard and Smart Education web pages.     |
+-----------------------------------------------------------------------------------+
                                         |
+-----------------------------------------------------------------------------------+
| PHASE 4: LOW-BANDWIDTH PWA & FINAL TESTING (Months 7 - 8)                         |
| - Implement IndexedDB offline queue and chunked tus upload protocol.              |
| - Conduct WCAG accessibility audits, security testing, and demo polish.           |
| - Final SIH Submission by 30 September 2026.                                      |
+-----------------------------------------------------------------------------------+

19. Testing & Validation Plan

• Functional Testing: Automated Cypress E2E tests verifying paper upload, search filters, spatial bounding box queries, and admin approval workflows [INFERENCE].
• AI Quality & Hallucination Testing: Evaluate 30 generated PR drafts against original paper texts. Enforce strict prompt constraints: "Include only facts present in the text snippet; do not extrapolate numerical results." Verify zero false metrics [INFERENCE].
• Low-Network / Offline Testing: Test PWA performance under offline mode and high latency (>1000ms) to ensure no data loss during interrupted uploads [INFERENCE].
• Security Audit: Vulnerability scanning with OWASP ZAP; verify role-based authorization ensuring standard users cannot access embargoed datasets or administrative review tools [INFERENCE].

20. Risk Register

21. Cost & Scalability

Hackathon Stage

• Hosting & Compute: $0 (Deployed on Vercel free tier for Next.js frontend, Render/Railway free tier for FastAPI, Supabase free PostgreSQL/PostGIS) [INFERENCE].
• AI API Cost: ~$5–$10 total usage using OpenAI GPT-4o-mini / Llama-3 API credits [INFERENCE].
• Total Hackathon Expense: Free / Minimal (< ₹1,000) [INFERENCE].

Production Deployment (MoES / NCPOR Data Center)

• Infrastructure: On-premise NIC/MoES cloud server deployment (2x Linux App Servers, 1x Database Server, 1x MinIO Storage Node) [INFERENCE].
• Storage Expansion: Scalable MinIO S3-compatible storage cluster (~20 TB initial storage for high-res video and datasets) [INFERENCE].
• AI Execution: Self-hosted open-source Llama-3-70B model on internal GPU server (e.g., NVIDIA A10G) to eliminate per-token cloud API costs and guarantee data privacy [INFERENCE].

22. Differentiation

23. Future Scope

1. Maitri-II IoT Real-Time Telemetry Integration: Direct API integration with the upcoming Maitri-II station [RESEARCHED] to stream live station weather data and ambient environmental metrics directly to the public portal [INFERENCE].
2. 3D/VR Virtual Station Expeditions: WebXR-based 3D virtual walkthroughs of Bharati and Maitri research stations for school students across India [INFERENCE].
3. Automated Podcast / Audio Explainers: Text-to-speech AI generation producing short bilingual audio podcasts summarizing recent polar science discoveries for educational broadcast [INFERENCE].

24. Assumptions & Open Items Log

Consolidated Assumptions Log

• [ASSUMPTION 1]: NCPOR outreach officers will perform final manual verification on all AI-generated social media content before public publishing.
• [ASSUMPTION 2]: Public educational users have basic modern web browser access supporting WebGL for 3D station map visualizations.
• [ASSUMPTION 3]: Legacy expedition reports available as physical books at the NCPOR library can be scanned to PDF format for OCR processing.

Consolidated Open Items & Validation Needed Log

• [NEEDS VALIDATION 1]: Exact network firewall rules and internal API access policies governing self-hosted deployments within MoES/NIC servers.
• [NEEDS VALIDATION 2]: Availability of official API keys for direct institutional social media publishing (X Enterprise API, LinkedIn Organization API).

Final Recommendation

Build PolarIS using Approach 3 (Decoupled FAIR Knowledge Repository + Multimodal AI Content Studio + Low-Bandwidth Field Client). This solution directly satisfies all explicit requirements of SIH26063 while offering maximum real-world value to NCPOR and MoES [INFERENCE].

Recommended Stack

• Frontend: Next.js 14, TypeScript, Tailwind CSS, Shadcn UI, Leaflet.js
• Backend: FastAPI (Python 3.12), FFmpeg, PyMuPDF, LangChain
• Database & Storage: PostgreSQL + PostGIS + pgvector, MinIO Object Storage, Redis
• AI/ML: Llama-3 / GPT-4o-mini (Summarization & PR), CLIP (Visual tagging)

Biggest Dependency

Access to structured historical expedition reports, publications, and sample datasets from NCPOR archives to populate the initial repository [INFERENCE].

Biggest Risk

AI generation producing subtle scientific inaccuracies or hallucinations in public press releases. (Mitigated via mandatory Human-in-the-Loop review) [INFERENCE].

Biggest Technical Challenge

Implementing seamless offline-first resumeable file uploads (tus.io) over high-latency (~800ms) satellite connections for polar station field scientists [INFERENCE].

First 5 Actions

1. Scrape and process 30 open-access NCPOR publications and ISEA expedition reports from public sources to construct a realistic demo seed database [INFERENCE].
2. Initialize the Next.js 14 frontend repository with Tailwind CSS and Shadcn UI components [INFERENCE].
3. Set up the PostgreSQL + PostGIS database schema and MinIO local bucket storage using Docker Compose [INFERENCE].
4. Build the FastAPI AI Content Studio pipeline using LangChain to test automated paper-to-tweet/LinkedIn conversion [INFERENCE].
5. Implement the Leaflet.js spatial dataset search interface with coordinates for Bharati, Maitri, and Himadri stations [INFERENCE].

Validation Before Coding

Verify that public NCPOR publication abstracts and expedition reports can be successfully parsed into clean Markdown/text using PyMuPDF [INFERENCE].

Buildability Score /100

92 / 100 (Highly feasible for a skilled 4–6 student team within the hackathon timeframe using modern Python/Next.js frameworks and open-source models) [INFERENCE].

Confidence

High (Supported by direct research on NCPOR operational infrastructure, NPDC portals, international polar metadata standards, and robust open-source software libraries) [INFERENCE].

Research Sources

1. Title: National Centre for Polar and Ocean Research (NCPOR) Official Website & Portal Directory
   URL: https://ncpor.res.in
   Source Type: Official Government Portal (MoES, Govt. of India)
   Supported Claim: NCPOR operational mandate, research stations (Maitri, Bharati, Himadri), library archives, and news announcements [PS] / [RESEARCHED].
   Reliability: High (Official Government Entity)

2. Title: Indian National Polar Data Center (NPDC) About & Data Policy
   URL: https://npdc.ncpor.res.in/npdc/mainmenu_home.action?ref_id=REF-30169&main_menu_id=5&main_menu_name=About
   Source Type: Official Dataset Repository Documentation
   Supported Claim: Functions, metadata flow, and open-data policies of India's National Polar Data Center established in 2013 [RESEARCHED].
   Reliability: High (Official Repository)

3. Title: NCPOR Significant Institutional & Technical Achievements
   URL: https://ncpor.res.in/antarcticas/display/369-significant-achievements-
   Source Type: Official Institutional Records
   Supported Claim: SATCOM satellite facility details (4 MHz link at Maitri), C-band data links at Bharati station via NRSC Hyderabad, and station infrastructure details [RESEARCHED].
   Reliability: High (Official Records)

4. Title: POLDER Federated Search & Schema.org Polar Metadata Documentation
   URL: https://polder-crew.github.io/Federated-Search-Documentation/intro.html
   Source Type: Academic / International Standards Repository
   Supported Claim: International best practice of using embedded Schema.org/Dataset JSON-LD metadata markup for global polar data discovery [RESEARCHED].
   Reliability: High (International Scientific Standards Body)

5. Title: ISO 19115 Geographic Information Metadata Standard Overview
   URL: https://www.dcc.ac.uk/resources/metadata-standards/iso-19115
   Source Type: Standards Organization Directory
   Supported Claim: ISO 19115 as the standard spatial metadata schema for marine and polar environmental datasets [RESEARCHED].
   Reliability: High (Digital Curation Centre / ISO)