Ranks Bhopal’s seven visarjan ghats by water-quality risk and allocates aeration units, testing kits and cleaning crews to the ghats that need them.
The problem
7 ghats. 9 days until Durga Visarjan. The BMC officer deciding where to send aeration units, cleaning crews, and testing kits has 1,524 readings buried across three 90–120 page MPPCB PDFs. Nobody reads them. Deployment decisions are made from memory.
How it works
Jal Prahari ranks all 7 Bhopal visarjan ghats by water-quality risk, scored against CPCB Class B bathing limits (DO ≥ 5, BOD ≤ 3, pH 6.5–8.5), weighted by recency and festival type, with every number page-linked to the source PDF.
The kill feature: Resource Allocator, enter how many aeration units, testing kits, and cleaning crews you have. The tool allocates each one to the ghat where it's actually needed, based on what that resource fixes (aeration follows oxygen risk, cleaning follows BOD risk). Shortfalls are named, not hidden. Ghats with missing data get testing first — crews are held back until results exist.
One-click bilingual deployment brief (English + Hindi), ready to print or WhatsApp. Built-in bilingual Q&A answers in English or Hindi — no AI required. Upload next year's CSV and everything recalculates with all-or-nothing validation. 38 automated tests. Works fully offline, zero API keys needed.
How Claude was used
Claude Code (Opus 5.5) built the entire codebase, server, risk scoring engine, 435-line bilingual Q&A engine, deployment brief generator, resource allocator, upload system with validation, UI with capsule charts, and all 38 tests. In the product itself, Claude is the optional "writer" AI: it can reword the deployment brief in natural Hindi/English, but every number it writes is verified against the coded fact sheet. If a number doesn't match source data, that field auto-falls back to the built-in template. AI never ranks, scores, or checks limits — that's all code in lib/risk.js.
Team
Jal Prahari