Product build2025Live

Review My Hostel

A real product shipped with agents as the primary engine

Live site

Stack

Frontend

Next.js 16 / React 19 student app
Angular admin panel
TypeScript
Tailwind CSS
Zod validation

Backend & Data

Supabase PostgreSQL with row-level security
Supabase Auth
Edge Functions for admin login, analytics and FAQ
OpenAPI contract
MCP server over the same data (TypeScript, Zod)

How it works

A student opens the Next.js app; signed in through Supabase Auth, they browse and submit reviews straight to Postgres under row-level security. Admins work from an Angular panel whose login, analytics and FAQ calls go through Edge Functions behind an OpenAPI contract, and moderation decides what goes live. An MCP server reads the same data for agent tooling.

Problem

Could agent-assisted development carry a product all the way to production — auth, data model, review flows, deployment — rather than stopping at a demo? This build was the test, with a live site as the pass condition.

What I built

A Next.js student app for browsing, reviewing and submitting, talking to Postgres directly under row-level security.
An Angular admin panel for moderation and analytics, with its login, analytics and FAQ calls behind Edge Functions and an OpenAPI contract.
Auth and data model on Supabase, with migrations checked in.
A read-only MCP server exposing the same data as typed, Zod-validated tools.
Deployed and live in production.

Technical decisions & trade-offs

A managed backend for auth and data, to keep the moving-parts count low while the workflow itself was the experiment.
Agents as the primary development engine, with human architecture and review as the quality gate.
Scope held deliberately tight — a complete small product beats an incomplete big one.

Outcome

Live in production. It showed that agent-assisted development reaches production quality when paired with real engineering judgment — a practice since carried into team workflows.