AI tools can write code fast, but nobody checks it. The same agent builds, tests and approves its own work, and when information is missing it guesses.

TayDau Force is a prototype built for the Alibaba/Qoder hackathon with Daud Sheraz. You describe a product in plain language, and a team of seven specialist AI agents takes it from idea to working software: requirements, planning, design, architecture, code, review and QA.

Each agent has one job. The agent that writes the code never approves it. A separate reviewer checks it, and a separate QA agent derives tests from the requirements. If a business detail is missing, the system asks the client instead of assuming it. Every requirement is marked as confirmed, proposed or assumed, so it’s clear what is fact and what is a guess.

The hardest problem was reliability. Free AI APIs hit rate limits and timeouts, which broke full runs halfway. I built a router that checks each model’s health and quota before sending a request, with a local fallback model if the cloud providers fail. Another bug showed a cleaning-services project rendered with an old portfolio design. I fixed that by isolating each project’s context and validating every design before use.

Stack: React, TypeScript, Vite, Tailwind, Node.js, Express, PostgreSQL, multi-provider AI gateway, Docker.

Live demo: https://taydau-force.vercel.app/