KSTM AI — Custom RAG SaaS

Architecture & System Design

A three-service polyglot monorepo consisting of a Next.js studio, a NestJS + Prisma API gateway (handling JWT auth, credit metering, and Stripe webhooks), and a Python FastAPI RAG microservice. The ingestion pipeline parses PDFs, CSVs, JSON, and JS-rendered URLs (via headless Chrome + BeautifulSoup) with MD5 caching, embeds chunks into isolated per-bot Neo4j vector indexes (`text-embedding-004`), and synthesizes multi-turn responses with Redis chat memory and Gemini 2.0.