retro-satquery

byDivya Rajput

# SATQUERY X — Build Complete Production Project You are a senior AI Architect, ISRO Remote Sensing Engineer, GIS Expert, Full Stack Developer, and ML Engineer. Build a complete production-ready Smart India Hackathon project named **SatQuery X** for **ISRO Problem Statement 26167 – SatQuery AI**. ## Objective Develop an **Agentic Vision-Language Assistant** that analyzes satellite imagery using natural language. The system must accept: * Single Optical / Multispectral image * Single SAR image * Optical + SAR image pair * Bi-temporal image pair Supported formats: * GeoTIFF * TIFF * PNG/JPEG (only for benchmark demo) The application must automatically select the correct AI workflow according to the user query. ## Mandatory Features Implement all ISRO requirements: 1. Image Upload 2. GeoTIFF validation 3. CRS checking 4. Resolution compatibility 5. Metadata extraction 6. Single Image VQA 7. Caption generation 8. Region grounding 9. Bi-temporal change detection 10. Optical + SAR fusion 11. Agentic Planner 12. Confidence score 13. Execution summary 14. Downloadable PDF report ## Innovation Do NOT use Planner Agent as novelty. Create an original module named **GeoReason Engine**. Responsibilities: * Cross-modal verification * Optical vs SAR contradiction detection * Temporal consistency verification * Explainable confidence score * Evidence fusion ## Architecture Frontend: * React * Next.js * TypeScript * Tailwind CSS * Leaflet Map Backend: * Python * FastAPI * Async APIs AI Models: * Llama 3.1 → Planner Agent * RemoteCLIP → VQA & Caption * GroundingDINO → Region Grounding * ChangeCLIP → Change Detection * CDVQA → Change Question Answering * SEN12MS Encoder → Optical + SAR Fusion GIS: * Rasterio * GDAL * GeoPandas Database: * PostgreSQL * PostGIS Deployment: * Docker * Ubuntu Linux ## Folder Structure satquery-x/ ├── frontend/ ├── backend/ ├── ai_models/ ├── datasets/ ├── database/ ├── docker/ └── docs/ Generate every folder with complete code. ## Frontend Pages Create fully responsive UI. Pages: * Landing * Upload * Dashboard * Report * Settings Dashboard contains: Left Panel * Satellite Map * Layer Control * Image Viewer Right Panel * AI Chat * Confidence Meter * Detected Objects Bottom * Execution Summary * Models Used * Download Report ## Backend APIs Create production REST APIs. POST /upload POST /validate POST /planner POST /analyze POST /report GET /health Use proper Pydantic models. ## Validation Agent Automatically detect: * CRS * Sensor * Resolution * Bands * Acquisition Date Reject incompatible images. ## Planner Agent Logic Classify user intent. Example: Query: Highlight water body Return: Task = Grounding Models = GroundingDINO + RemoteCLIP Query: What changed between 2022 and 2026? Return: Task = Change Detection Models = ChangeCLIP + CDVQA Do not expose reasoning. Only output execution trace. ## AI Pipelines Pipeline 1 Single Image Input Optical Tasks * VQA * Caption * Grounding Pipeline 2 Bi-temporal Tasks * Change map * Change description * CDVQA Pipeline 3 Optical + SAR Tasks * Flood detection * Water mapping * Built-up verification ## GeoReason Engine Algorithm: Collect outputs from all models Compare predictions Check metadata Remove contradictory detections Generate: * Final answer * Confidence % * Evidence summary Example: Optical = Water +22% SAR = Water +8% Final = 8% Confidence = 95.4% ## PDF Report Generate professional report containing: * User Query * Images * Highlighted Map * AI Answer * Statistics * Confidence * Models Used * Timestamp ## UI Theme Modern ISRO inspired. Colors: Blue White Dark Navy Cards with rounded corners. Responsive for desktop and laptop. ## Deliverables Generate: * Complete source code * Docker setup * README * Requirements.txt * Package.json * SQL schema * API documentation * Installation guide * Testing guide * SIH presentation content ## Coding Rules Write clean architecture. Use comments. Follow SOLID principles. Use reusable components. No dummy code. Every module must be fully functional. Generate the project **step by step**, beginning with Part 1 (Project Setup), then Frontend, Backend, AI Integration, Database, Docker, and finally Testing.