Team Name: Divya Drishti (दिव्य दृष्टि)
Target Challenge: Problem Statement 26167 — SatQuery AI
Sponsoring Agency: Indian Space Research Organisation (ISRO) — Space Applications Centre (SAC)
Team Leader: Tanmay Shah (230020107079)
SatQuery AI is an operational-grade Geospatial Intelligence & Remote Sensing Vision-Language Studio. Designed to move beyond generic LLM chatbots and "AI wrappers," SatQuery AI provides an interactive, map-centric workbench with native raster ingestion, multi-spectral band processing, and pixel-grounded geospatial reasoning.
- Map-Centric Studio Canvas: 60fps MapLibre GL vector and satellite globe with multi-layer overlays (optical, SAR, thermal, and ND-indices).
- Native Geospatial Ingestion: Parses Cloud-Optimized GeoTIFFs (COGs), Sentinel-2 MSI, Sentinel-1 SAR (GRD), and Cartosat imagery with preserved UTM/WGS84 geotransforms.
- Vision-Language Reasoning: Specialist VLM pipeline for earth observation using RemoteCLIP, VRSBench evaluation, and CDVQA benchmarks.
- Automated Change Detection: Multi-temporal change analysis with pixel-level difference masks and sector timeline tracking.
- Auditable Execution Trace: Transparent inspection drawer displaying model routing, confidence metrics, latency, and tensor dimensions.
- Operational Intelligence Export: Export detected AOIs, vector footprints, and mission reports directly into GeoJSON, Shapefile, and PDF briefings.
├── backend/ # FastAPI high-performance backend
│ ├── app/
│ │ ├── agent/ # VLM agent router, action planner & evidence builder
│ │ ├── api/ # REST & WebSocket endpoints (query, STAC, benchmark, upload)
│ │ ├── engine/ # GeoTIFF parser, SAR fusion, change detector
│ │ └── models/ # RemoteCLIP, BigEarthNet adapters, model registry
│ └── tests/ # Comprehensive test suite & API contracts
├── frontend/ # React 19 + TypeScript + Vite + Tailwind CSS
│ ├── src/
│ │ ├── components/ # MapCanvas, OperationalPanels, SwipeCurtain, Globe
│ │ ├── pages/ # Studio, AuditStudio, Benchmarks
│ │ └── stores/ # Zustand state stores (query, UI, layers)
├── data/
│ ├── benchmarks/ # CDVQA and VRSBench benchmark ground truths
│ ├── sample/ # Sample GeoTIFFs, SAR products, and sector assets
│ └── weights/ # Model weights & adapters (large weights git-ignored)
├── sih-info/ # Comprehensive SIH 2026 Master Dossier & slides
│ ├── README.md # Master dossier quick start & index
│ ├── presentations/ # Official PPTX decks & presentation templates
│ ├── assets/ # Presentation images & diagrams
│ ├── diagrams/ # Generated diagrams & howlers extraction
│ ├── 07-satquery-production-architecture.md
│ ├── 08-satquery-datasets-and-tech-ecosystem.md
│ └── 09-ui-ux-design-specification.md
├── scripts/ # Setup, start, and utility scripts
│ ├── setup.py # One-command setup (Mac/Windows/Linux)
│ ├── start.sh / start.bat # One-line server start
│ ├── run_app.sh # Launch both backend + frontend
│ └── generate_presentation*.py # Presentation generation scripts
├── docs/ # Audit reports, remediation logs, limitations
├── deploy/ # Docker & deployment configs
│ ├── Dockerfile # Containerized deployment
│ ├── docker-compose.yml # Local stack orchestration
│ └── nginx.conf # Reverse proxy & static serving config
├── setup.py # Root-level project setup
└── README.md # This file
python setup.pyThis creates a Python venv, installs all backend dependencies, sets up backend/.env, installs frontend npm packages, and checks for Ollama. Requires Python 3.10+ and Node.js 18+ pre-installed.
# macOS / Linux / WSL:
bash scripts/start.sh
# Windows:
scripts\start.batThe backend runs its Vision-Language narrative synthesis on a local Qwen2.5:3B model served by Ollama. Install Ollama, then start it and pull the model:
# Install Ollama (macOS via Homebrew, or download from https://ollama.ai)
brew install ollama
# Start the Ollama service
ollama serve &
# Download the model the backend expects (qwen2.5:3b by default)
ollama pull qwen2.5:3bThe exact model name can be overridden with OLLAMA_MODEL in backend/.env (e.g. qwen2.5:7b). Without Ollama the app still runs in fallback (DSP-only) mode, but local neural reasoning is disabled.
cd backend
python3 -m venv .venv
source .venv/bin/activate # Windows: .venv\Scripts\activate
pip install -r requirements.txt
uvicorn app.main:app --reload --host 0.0.0.0 --port 8000cd frontend
npm install
npm run devFull technical documentation, architectural specifications, and presentation slide decks are available in the sih-info/ directory:
- Master Dossier Index
- Official Problem Statement 26167 Details
- System Architecture Specification
- UI/UX & Map Experience Design
- Datasets & Remote Sensing Models
- Team Roster & Compliance
Developed for the Smart India Hackathon (SIH) 2026 under the ISRO / SAC challenge. All rights reserved by Team Divya Drishti.

