ai-chatbot-python-huggingface-completed

This project is a lightweight, CPU-friendly AI chatbot built with an open-source Hugging Face model. It is designed as a foundation for business use cases, incorporating knowledge-base lookup, a web interface, guardrails, and persistent memory. The chatbot is powered by Qwen2.5-0.5B-Instruct, which runs on CPU without requiring a GPU. It maintains conversation history so the bot remembers context within a session, and includes a model upgrade helper to switch to a bigger model when a GPU is available. The system features a simple knowledge base search that uses keyword matching against company facts, which are injected into the prompt for more relevant answers. A web interface is provided via Gradio, with a shareable public link. Basic guardrails block inappropriate messages before they reach the model, and persistent memory saves and reloads conversations per user with SQLite, so chats survive restarts. The tech stack includes Python 3.12, Transformers, PyTorch, Gradio, and SQLite. Known limitations include that the default model is small and can sometimes give generic or repetitive answers, knowledge base search uses simple keyword matching rather than semantic search, and guardrails use basic substring matching and are not a complete content-safety solution.