Case study
Web Application
BuildWise AI
Drop a pin anywhere and get a full civil-engineering feasibility report in seconds — MCDA land suitability, IS 1893 seismic and flood stress, CPWD cost benchmarks, Vastu orientation, and a generated architectural blueprint, compiled into a print-grade PDF dossier.
The problem
Site feasibility is slow, expensive, and fragmented. Before buying land, a builder has to commission separate surveys for land suitability, flood risk, seismic zoning, budget, and orientation — each with its own consultant, timeline, and invoice. Small developers and self-builders usually skip the analysis entirely and discover the problems after purchase.
The solution
Built a geo-based planning platform that runs the whole feasibility stack from one map pin: MCDA land-suitability scoring, IS 1893 seismic and flood stress, CPWD cost estimation, Vastu orientation, and a CSP-solved generative floor plan. A deterministic civil core keeps the maths auditable, while Groq/Gemini adapters add reasoning with a zero-key offline fallback.
Outcome
Produces a print-grade PDF dossier — executive scorecard, vector blueprint, and room schedule — in seconds. Runs entirely on open data (OpenFreeMap, Esri, USGS, Open-Meteo, Sentinel-2), so there are zero paid APIs. Validated by a 236-test backend suite, a 145-test placement regression suite, and a 9-viewport Puppeteer UI audit.
What I learned
Constraint solving is where architectural plausibility lives or dies — early layouts were geometrically valid but unlivable until topological invariants (door adjacency, wet-zone grouping, daylight access) became hard constraints rather than heuristics. Also learned to keep the deterministic civil core strictly separate from the LLM layer: numbers must never come from a model. Would add per-region building-code profiles next so the engine is not tied to Indian standards.
Stack
- React 19
- TypeScript
- FastAPI
- Python
- MapLibre GL
- Groq
- Gemini
- ReportLab