Ideas for useful public-good tools I've thought about but haven't built. If one of these resonates and you'd like to build it (or help), reach out: raghavsyz@gmail.com.
The idea
A map of India showing where garbage/waste is illegally dumped, crowdsourced and/or detected from imagery, so NGOs and local government can prioritize cleanup — with particular attention to tourist and heritage zones.
Brainstorming notes
Data sources
Free satellite imagery (Sentinel-2, ~10m resolution; ISRO Bhuvan/Copernicus) is too coarse to spot street-level dump sites. Commercial high-resolution imagery (Planet Labs, Maxar, SkyWatch — roughly 0.3-3m resolution) would be needed for real detection; some offer academic/non-commercial tiers worth checking. Google Earth Engine API is useful for batch analytics over historical imagery.
Possible stack
A computer vision model (YOLO or Segment Anything, fine-tuned on a dataset like TACO or custom-annotated satellite waste imagery) to detect dump sites → a Python backend (FastAPI/Flask) → PostGIS for geospatial storage and queries (e.g. "find all sites within 5km of a tourist landmark") → Mapbox GL JS or Leaflet.js for the map frontend.
Features worth considering
Tourist-zone priority scoring, by overlaying detections with OpenStreetMap heritage/hotel data. Crowdsourced human-in-the-loop verification, where citizens or NGO volunteers confirm detections with a geotagged photo. Automated report/alert generation formatted for municipal bodies (e.g. MCD, BBMP).
Open question
Whether to start with a simple crowdsourced manual-pin prototype first, or attempt satellite-based automatic detection from the start. No strong opinion yet — open to discussing with anyone interested in building this.
The idea
A platform where companies list real R&D problems/open questions they have, and researchers/professors (starting with IITs) can submit proposals or applications describing how their lab would approach it. Companies then choose which lab to collaborate with. The motivation is to increase student interest in non-CS fields by exposing them to real industry problem statements and collaborations, rather than only abstract coursework.
Brainstorming notes
Core flow
Company posts a problem statement (scope, domain, constraints, what they're offering — funding, data, access, internships). Labs/professors browse and submit a proposal (approach, timeline, team). Company reviews proposals and selects a lab to collaborate with.
Open questions
How to vet companies and problem statements so they're genuinely open (not just disguised outsourcing). How selection/matching should work — direct company choice vs. some lightweight review process. Whether to start narrow (one institute, one domain) before generalizing across IITs and disciplines. No strong opinion yet — open to discussing with anyone interested in building this.
The idea
Assam has seen severe recurring flooding (notably in 2026, across Upper Assam districts like Sivasagar, Charaideo, Jorhat, and Golaghat), affecting hundreds of thousands of people. A system combining satellite imagery, river gauge data, weather forecasts, and communications infrastructure could provide village-level flood forecasts and disaster planning — technology that could save lives directly in a region prone to monsoon flooding, cloudbursts, and Brahmaputra tributary overflow.
Existing tools/inspiration
This space isn't empty — worth studying before building anything new:
Possible angle
Given the above, the highest-value contribution may not be a new global forecasting model (Google Flood Hub already does that reasonably well) but a last-mile layer specific to Northeast India: finer village-level resolution, integration with local communications infrastructure for actually reaching at-risk villages, and combining official CWC data with satellite-based verification.
Open questions
Whether to build on top of/extend an existing system (e.g. contribute Northeast India coverage/data to Google Flood Hub) versus building an independent tool. How to handle last-mile alerting in areas with limited connectivity. No strong opinion yet — open to discussing with anyone interested, including on the AI/ML modeling side (remote sensing, flood forecasting) or the on-the-ground disaster response side.