Problem
Construction site managers in India report snags (defects, delays, materials issues) over WhatsApp — text, voice notes, photos. The data never makes it into a system of record, so progress reviews are reconstructed from screenshots. ForeSites is the system of record: a WhatsApp-first ingestion pipeline plus a dashboard that lets the developer query and assign across sites.
My contributions
NL-to-SQL agent (AWS Lambda · Gemini)
Built a Python AWS Lambda agent (Gemini API) that performs streaming NL-to-SQL over a 9-table Postgres schema. Per-session memory is persisted across turns; session-ownership enforcement keeps each conversation scoped to a single project so cross-tenant leaks are impossible by construction.
WhatsApp ingestion service (Node · Express · S3 · Supabase)
Built the ingestion service: WhatsApp Business API webhook → Express handler → S3 for media → Supabase (Postgres) for the structured snag record. Voice notes are transcribed, images are OCR’d, text snags are written directly. Each snag carries a project ID derived from the sender’s project membership.
Dashboard + schema
Redesigned the React dashboard frontend; authored the relational schema and the snag-assignment workflow (assignee, status, SLA, audit trail).
Status
Pilot stage with a Pune real-estate developer. The platform is producing usable snag records and the agent answers project queries in production. Next: cost-aware routing for the agent and a structured snag-template layer above the freeform ingestion.