bnao – AI Construction Guides from Site Photos

bnao ("banao", Hindi for build) is an AI SaaS I designed and built end to end. A builder uploads photos of a plot and answers a short questionnaire; bnao returns a step-by-step 3D build guide their crew can follow on site, with quantities, costs and checklists, in English and Hindi.

What I built
• Multi-step AI job pipeline on the Claude API, with its own evaluation harness and per-job token cost tracking. By design the model never outputs prices or geometry directly: it plans, and deterministic code computes.
• Interactive 3D guides with Three.js / react-three-fiber, step by step, with furniture, fittings and services placed room by room.
• Supabase backend: Postgres with row-level security, phone-OTP auth and private storage. Every one of the 12 migrations is tested against real Postgres in CI.
• Billing with two providers (Razorpay and Paddle), GST-compliant invoicing, money stored as integer minor units, and a daily billing cron.
• Teams, invites, admin tools and public share links, protected by Cloudflare Turnstile.

Engineering quality
• Vitest unit tests, database policy tests and Playwright end-to-end tests with axe accessibility checks.
• Lighthouse CI budgets: accessibility, best practices and SEO ≥ 0.95, layout shift under 0.1.
• GitHub Actions for CI and database deploys; changes merged through pull requests.