saket kunwar

BlindSpot is an auditable neuro-symbolic disaster intelligence system built for the moments when clouds, smoke, and revisit gaps blind orbital satellites. Built on the strict law that "Muse proposes, the ledger disposes," BlindSpot isolates Meta’s Muse models (muse-spark-1.2, 1.3, and muse-wonder-1.0) from names, dates, and web access to frame rival physical mechanisms from anonymized geometry, while a deterministic Qualitative Spatio-Temporal Reasoning (QSTR) kernel—using RCC-8 topology, Allen’s interval algebra, and directional tiles—scores those hypotheses across the sensing gap. Validated on the August 2026 Tibet-Nepal secondary dam burst (scoring 0.9426 and triaging 171 bridges across a 172.8 km danger zone) and ongoing Kalimantan peat fires (predicting underground reignition at 0.8425 and monitoring live FIRMS detections via an Autonomous Collaborative Agent with four-tier cross-continental memory and a self-evolving Skeptic loop), every byte is protected by 25 hardened coding-time and execution-time verification gates. Explore the interactive demos at: https://blindspot-hazel.vercel.app/