← cd ../work

// Marketplace · AI engineering

Peer-to-peer car rental marketplace

A full product built end to end with an agent-driven workflow on Claude Code, with 400+ pull requests merged in its first four weeks.

company
Personal project
role
Sole engineer
period
2026
status
Personal project

What it is

A personal project built in my spare time, not a commercial venture. It’s an exercise in shipping a complete product end to end: a Next.js web app, a Nest.js API (PostgreSQL, Redis, Stripe Connect), and native iOS (SwiftUI, TCA) and Android (Kotlin, Compose) apps, all in one monorepo.

An agent-driven development workflow

I designed a development workflow on Claude Code where custom skills take a task from a one-line description to a merged pull request: spec, implementation, review, QA and merge. Dedicated programmer, reviewer and QA subagents run each step, with platform-specific reviewers for iOS and Android.

The result was more than 400 pull requests merged in the first four weeks.

Making parallel AI work safe

Each task runs in its own git worktree with dedicated databases, storage buckets and ports. Several agent sessions can build, test and merge at the same time without colliding.

AI in the product

I built an in-app AI support agent on the Claude API:

  • Tool use over the customer’s own bookings
  • The full product knowledge base served from a shared prompt cache instead of retrieval
  • Streamed replies and human hand-off
  • A provider-agnostic model layer (Claude, Bedrock, Nova), backed by an eval suite