RAG-Based Knowledge Retrieval System (TypeScript/LangChain)

Local vector store RAG demo using TypeScript and LangChain.

• ingest command to chunk and index source documents in knowledge base
• chat command for interactive Q&A over retrieved context from knowledge base
• Azure OpenAI for embeddings and chat
• PostgreSQL + pgvector for retrieval
• Chunking, retrieval, and context combination are delegated to LangChain primitives.

Requirements

• Node.js 22+
• Docker
• Azure OpenAI deployment(s)
• Managed identity (recommended) or API key auth

Configuration

All runtime settings live in config.yml:

• Azure endpoint, API version, deployment names
• Vector store connection and table names
• Ingest chunking settings
• Chat retrieval settings and system prompt

Environment variables

Only one environment variable is optionally used:

• AZURE_OPENAI_API_KEY (only needed when not using managed identity)

Install

npm install

Start local vector store

docker compose up -d

Ingest documents

Provide one or more file paths to ingest:

npm run ingest -- ITM.pdf green-belts-v27.pdf style-guide-for-decisions.pdf

Chat with the knowledge base

npm run chat

Then ask questions in the prompt; type ctrl+d to exit.