// the find
alfredfrancis/ai-chatbot-framework
A python chatbot framework with Natural Language Understanding and Artificial Intelligence.
A self-hosted chatbot builder with a web admin UI for defining intents, entities, and conversation flows without writing code. It combines traditional ML-based NLU (scikit-learn, CRF, spaCy embeddings) with LLM zero-shot classification as a fallback. Aimed at teams who want a Dialogflow-like thing they can run themselves.
The dual NLU approach is practical — trained ML classifiers for known intents, zero-shot LLM for everything else, which keeps costs low for high-traffic bots. Docker Compose and Helm charts are both included, so you can go from laptop to Kubernetes without reinventing deployment. The intent/entity training UI with conversation logs is genuinely useful for non-ML people who need to iterate on bot behavior. FastAPI + Motor (async MongoDB) is a reasonable stack choice for a chatbot backend where latency matters.
The repo description and primary language tag say TypeScript/Python but the actual ML pipeline is squarely Python — the TypeScript label is just the Next.js frontend, which will confuse people searching by language. Slack and WhatsApp integrations are listed as 'coming soon' with no timeline, and the RAG knowledge base is 'in development' — the feature list overpromises what's actually shipped. MongoDB as the only storage option is a hard dependency that rules out teams standardized on Postgres. Test coverage is thin: one test file for the dialogue manager with no apparent NLU pipeline tests, which is the hardest part to get right.