The ultimate AI product manager. ShipFlow orchestrates the entire software lifecycle from a raw idea to deployment-ready code.
- Next.js
- tRPC
- Prisma
- PostgreSQL
- Inngest
- Tailwind CSS
Problem
The gap between product ideation and engineering execution is massive. PRDs get ignored, QA reviews take days, and miscommunication causes endless delays in shipping features.
Approach
I built a comprehensive multi-tenant SaaS that acts as an intelligent bridge. It uses AI to challenge and refine product requirements, automatically generates engineering tasks, and runs a relentless autonomous QA review against GitHub Pull Requests.
Architecture
- Next.js App Router Monorepo powered by tRPC
- Inngest durable workflows for resilient AI code reviews
- Deep GitHub OAuth & Webhook integration for PR tracking
Results
- Fully automated the QA review loop directly inside GitHub
- Reduced PR friction by instantly catching requirement deviations
Architecture
tRPC Monorepo
Workflows
Durable
// code
shipflow-ai.ts
// AI PRD Generation via Inngest durable workflows
export const generatePrd = inngest.createFunction(
{ id: "generate-prd" },
{ event: "feature.analyzed" },
async ({ event, step }) => {
const prd = await step.run("query-llm", () =>
ai.generate({ prompt: buildPrdPrompt(event.data) })
);
await step.run("save-to-db", () => db.prd.create(prd));
return prd;
}
);lessons learned
- Inngest's durable execution ensures LLM timeouts or API failures don't drop the entire generation pipeline.
- Webhook-driven state machines for GitHub PRs are vastly superior to polling.