How I Built 4 Production AI Products in 6 Months

Description and use

Over six months, the author shipped four production AI products using a shared Next.js/NestJS stack, pgvector for RAG, and LangGraph for agent workflows. The products include a virtual receptionist, PDF search, AI support agent, and a freelancer hiring server.

Problem

Clinics, salons, and service businesses miss calls during busy hours; support teams answer repetitive questions; teams waste time searching PDFs; clients find it hard to vet freelancers.

Outcome

All four products are live, testable, and solve real problems, demonstrating a repeatable production‑first AI development workflow.

Workflow

• 1. Define product scope and data sources
• 2. Build Next.js frontend with tRPC/REST
• 3. Implement NestJS backend modules with DI and guards
• 4. Integrate pgvector for RAG and LangGraph for agent workflows
• 5. Add WebSocket for real‑time interactions
• 6. Deploy to production domain
• 7. Conduct weekly live demos and iterate

Evidence

Clinics, salons, and service businesses miss calls during busy hours. Those callers book elsewhere.

Source

Drop any PDF (even 40+ pages), ask a question, get a cited answer with page numbers.

Source

Reads your documentation once, then resolves tickets, checks order status, and escalates with full context.

Source

All 4 products share a production‑grade foundation: Next.js Frontend, NestJS Backend, pgvector, LangGraph, WebSocket.

Source