KnowledgeHub AI
Multi-tenant RAG SaaS where teams upload their documents and get grounded, cited answers. A NestJS API owns auth, organizations, ingestion, and retrieval; a TanStack Start app streams answers back over SSE.
- Hybrid retrieval: pgvector embeddings plus PostgreSQL full-text search, with query rewriting before generation.
- Every answer cites its source chunks, with an evaluation suite that checks citation validity.
- Organizations, roles, and invitations, plus per-tenant rate limits and usage quotas to keep LLM spend bounded.
- Presigned uploads to Cloudflare R2 and a background processing pipeline for extraction, chunking, and embedding.
NestJS · TanStack Start · PostgreSQL · pgvector · Prisma · OpenAI · Redis · Cloudflare R2 · Railway