RAG Bootcamp
Reference implementations from Vector Institute's RAG bootcamp that show common retrieval-augmented generation methodologies over web, document, SQL and cloud data sources using LangChain and LlamaIndex.
This repository collects the reference implementations used in Vector Institute's RAG Bootcamp, run from November 2024 to January 2025, demonstrating data ingestion, chunking, embeddings, vector databases, sparse and dense retrieval, and reranking with the Python LangChain and LlamaIndex libraries. Separate implementations cover web search for questions outside a model's training data, document search over unstructured collections, SQL search over a structured financial dataset, and cloud search over data held in AWS S3.
A PubMed question-answering implementation runs a full pipeline of ingestion, embeddings, vector storage, retrieval and reranking with a focus on evaluation metrics, and a separate RAG evaluation implementation applies the Ragas framework to build test sets and measure how well a retrieval pipeline is actually working. Setup uses Python 3.10 or later with uv for the virtual environment and dependencies, and a Jupyter kernel for running the notebooks.