Zambezi Voice RAG: A RAG System Starter Code

Zambezi Voice RAG is a hybrid search retrieval-augmented generation system built with Streamlit. It combines dense vector embeddings via mxbai-embed-large and sparse BM25 retrieval on a Pinecone index for accurate, context-aware answers. Users can query research documents conversationally, inspect retrieved chunks, and get streaming responses powered by Llama 3.1 via Ollama.

Key Features

• Hybrid search combining BM25 sparse and dense vector retrieval
• Streaming LLM responses powered by Llama 3.1 via Ollama
• Inspectable retrieved document chunks per query
• Persistent chat history within session

Tech Stack