DocuMind AI is an advanced Retrieval-Augmented Generation (RAG) conversational chatbot engineered to eliminate hallucinations when querying private, specialized organizational documents. Built on Google’s open-weight Gemma 4 (gemma-4-26b-a4b-it) and Sentence-Transformers, DocuMind AI upgrades traditional single-turn RAG into a production-grade multi-turn conversational assistant.

Unlike basic prototypes, DocuMind AI features a Dual-Tier Guardrail: an algorithmic vector relevance threshold that intercepts off-topic queries before making LLM calls, saving API costs and latency, followed by strict citation rules that attribute facts to source files and page numbers (e.g., (handbook.pdf, p. 2)). It ingests multiple document formats (PDF, Markdown, and TXT) and runs entirely on a free-tier Google Colab CPU without requiring expensive GPU servers.