DocMind AI: A conversational document question answering system

The growing volume of unstructured technical documentation in engineering, academic, and enterprise settings makes manual information retrieval slow and often incomplete. Generalpurpose large language models answer natural-language questions fluently, yet they tend to produce unsupported answers when queried about material outside their training data and cannot reason over documents supplied at runtime. DocMind AI addresses this gap with a conversational document question-answering system that applies the Retrieval-Augmented Generation (RAG) paradigm to ground language-model responses in user-supplied documents. Built with the LangChain framework, it accepts PDF, DOCX, and TXT files through a Streamlit interface, splits their content into overlapping text chunks, and encodes each chunk with the sentencetransformers/all-MiniLM-L6-v2 embedding model. The resulting vectors are stored in a persistent ChromaDB database and retrieved by cosine similarity at query time. Retrieved passages, together with prior conversation turns held in a buffer memory, are passed to a Groq-hosted LLaMA 3.3 70B versatile model through a conversational retrieval chain, enabling coherent multi-turn dialogue grounded in the source documents. The complete pipeline was implemented end-to-end and exercised on a systems-engineering reference corpus, producing contextually relevant, source-grounded answers across a range of question types. The system architecture, implementation details, and qualitative observations gathered during testing are described in the sections that follow.