Developer builds RAG system from scratch to answer questions from private documents

A developer shared a hands-on project building a Retrieval-Augmented Generation (RAG) system designed to answer questions strictly from the content of a company policy document. The system uses LangChain, sentence-transformers, a Chroma vector store, and Ollama's TinyLlama model to retrieve and generate context-grounded responses. Documents are split into 200-character chunks with 50-character overlap, embedded, and stored so the most relevant passages can be retrieved per query. The developer noted that answer quality depends not just on the language model, but also on chunking strategy, retrieval accuracy, and prompt design. The project is intended as a foundation for further experiments with multi-agent workflows and autonomous AI systems.