Loading...
Loading...


A full-stack records-and-visibility platform (internal codename techBuilder) for running an Indian construction company's daily field operations. Site managers, team heads (mistri) and drivers log simple end-of-day records in Hindi on their phones; those records roll up to an Owner dashboard with Excel export. It is deliberately a logbook, not project management or BIM. The monorepo is three npm workspaces — shared (frozen, versioned Zod contracts that both ends compile against), backend (NestJS + Drizzle + Postgres with row-level security, orgId on every row) and web (Next.js App Router, httpOnly-cookie auth, TanStack Query, shadcn/ui) — plus a Bubblewrap TWA that ships the same origin to the Play Store instead of a separate React Native app. Five roles (Owner, Site Manager, Team Head, Driver, Worker) with a frozen RBAC matrix; every backend route re-derives its scope from the database, so route guards in the UI are convenience, not security. Runs self-hosted on a single ap-south-1 EC2 box — local Postgres 18, nginx, pm2 — behind Cloudflare.
Role
Solo Full-stack Developer (architecture, backend, web, deployment)
Duration
2026
Indian construction SMBs run on paper diaries and WhatsApp. Owners have no reliable picture of what happened on site yesterday, and every existing tool is English-first project-management software aimed at far larger firms.
A Hindi-first logbook where field roles enter a handful of fields at the end of the day on their own phones, records roll up to the owner in real time, and the whole thing is small enough to run on one small server per deployment.
Live at techbuilder.pooniya.com with the full 5-role portal, an Android TWA build, and an onboarding model where a new company is provisioned by hand and then runs fully self-service.
Hindi-first UI — Hindi is the primary language of the product, not a translation layer bolted on later
Multi-tenant by construction: orgId on every row plus Postgres row-level security, so isolation is enforced in the database
5 roles (Owner / Site Manager / Team Head / Driver / Worker) with a frozen RBAC matrix and cascading account creation
shared/ workspace of versioned, frozen Zod contracts compiled by both backend and web — a contract change breaks the build, not production
httpOnly-cookie auth through a Next proxy — no bearer token ever touches client JS
Media pipeline with a pluggable storage driver (local by default, S3 optional) — every read passes the RLS + scope check before bytes are served
Feature-flagged per org (features.photos, voice notes) with defaults that reach already-stored configs without a backfill
Owner dashboard with ExcelJS export; offline indicator + PWA for patchy site connectivity
Shipped to Android as a Bubblewrap TWA wrapping the deployed origin instead of maintaining a second native codebase
Self-hosted end to end: single EC2 box with local Postgres, nginx and pm2 — no managed DB, no Vercel
~840 unit + integration tests across the three workspaces, with a theme-audit script as a build gate
Pivoted the frontend from Expo/React Native to a Next.js web portal after a day of native-tooling friction produced zero backend bugs — shipping the Android app later as a TWA was the cheaper path to the same place
Freezing contracts in a shared workspace turns every breaking API change into a compile error instead of a runtime 500
Next matches a dynamic [id] segment before an optional catch-all — which silently 404'd every named sub-page until a route-conflict test was written to fail on the shape
A feature flag added as a NEW key reaches already-stored org configs via .default(false); flipping an EXISTING default does not













