RAG API Support AI Chatbot
A conversational AI system that helps developers and support teams answer questions about APIs by retrieving relevant documentation and generating contextual, grounded responses with source citations. Users can ask natural language questions like "How do I authenticate?" or "Why am I getting a 401 error?" and receive accurate answers pulled directly from ingested API docs. Built with Django REST Framework, PostgreSQL with pgvector for vector similarity search, and Ollama for local LLM inference — no cloud API keys required.
Problem Statement
Developers frequently struggle navigating complex API documentation, jumping between docs, examples, and support tickets to find answers. There was a need for a conversational interface that centralizes API knowledge and provides grounded, cited responses.
My Approach
Built a RAG pipeline where API documentation is ingested, chunked semantically, and converted to embedding vectors stored in PostgreSQL with pgvector. User queries are embedded and matched via vector similarity search to retrieve the most relevant doc chunks, which are then compiled into a prompt for the LLM (Ollama) to generate grounded answers with source citations. The entire stack is containerized with Docker for easy deployment.
Key Outcomes
- RAG-based question-answering over custom API documentation with source citations
- Vector-based semantic search across document chunks using pgvector
- Conversation tracking with message history for multi-turn interactions
- Fully local LLM inference via Ollama — no cloud API dependency
- Dockerized stack for simplified one-command deployment
Tech Stack
Tags
Project Info
- Status
- Completed
- Category
- Personal
- Created
- 1 year ago
- Ended
- Aug 2024
Links
Related Projects
Sales Data Agent - Enterprise AI with Snowflake Intelligence
Currently leading development of an enterprise conversational AI agent that enables business users to query Snowflake sales data in plain English — no SQL required. Built on Snowflake Cortex Intelligence with RAG, Text-to-SQL, and AI Observability.