No framework in the middle
The retrieval pipeline is written directly — extract, embed, store in Pinecone, retrieve, answer — so every step can be inspected and tuned.
← all work
case study · Applied GenAI · 2026
RAG Support Agent
A support platform where each business uploads its own documents and an AI agent answers customers from them — in web chat and on WhatsApp.
the pipeline
- Ingestpdf-parse · mammoth · multerPDF and Word uploads, text extracted
- RetrieveEmbeddings + PineconeThe most relevant passages for each question
- AnswerGroqA reply grounded in those passages
- DeliverSocket.ioLive web chat and WhatsApp
- My role
- Sole developer
- For
- Personal project · code on GitHub
- Status
- Code on GitHub · code ↗
- Stack
- Next.js · TypeScript · Express · MongoDB · Pinecone · Groq · Socket.io
the problem
Support teams answer the same questions from policies scattered across PDFs and Word files. A general chatbot doesn't know a business's own rules; this platform answers from the documents each business gives it.
what I built
- Multi-tenant: each business has its own knowledge base and conversations
- Document upload in several formats — PDF (pdf-parse) and Word (mammoth)
- Embeddings stored and searched in Pinecone
- Answers generated with Groq
- Real-time chat over Socket.io, on the web and on WhatsApp
- Next.js frontend; Express + MongoDB backend with JWT auth
decisions that mattered
One knowledge base per business
Each tenant's documents and conversations are kept separate from every other tenant's.
Live by default
Messages travel over WebSockets (Socket.io), so the chat updates without refreshing.
result
- Code on GitHub.
- The Ask console on this site uses the same idea: answers grounded in documents, with the sources shown.