Team Sudo Solve | SIH26167 Satquery| Demo recording Walkthrough

SatQuery AI — SIH26167 · Team sudo solve · Smart India Hackathon 2026 An interactive vision-language assistant for multimodal remote-sensing image analysis. Ask plain-English questions about satellite imagery and get auditable answers — every number comes from a deterministic tool, never from the model's imagination. Demo covers all three input modes: • Single optical image (Sentinel-2, Sundarbans) — water extent measured by NDWI, overlaid and quantified in km² • Bi-temporal pair (LEVIR-CD) — change detection with area and direction analysis; fields that can't be measured are withheld, not invented • Co-registered optical + SAR pair (Sentinel-1 GRD + Sentinel-2) — calibrated backscatter analysis with cross-sensor agreement Architecture: two open-weight Qwen3-VL models (a frozen narrator and an RSVQA-adapted answer model), an agentic router with a deterministic fusion planner, and a typed evidence packet behind every claim. Runs fully offline on a laptop GPU. No commercial vision APIs. PS: SIH26167 | Theme: Space Technology | Category: Software