How lowkey-one/melp_bot works
Project type and purpose
• This is an open-source conversational AI assistant built on the Rasa framework.
• It handles multiple domains (calendar, calls, chat, drive, team management, network queries, user profiles) via NLU, Core (dialogue management), and custom actions.
• It’s containerized for deployment (Docker, docker-compose).
File structure
Top level
• actions/ – Python action server code (actions.py)
• data/ – Rasa training data: NLU, rules, stories
• domain/ – Rasa domain definitions per feature
• models/ – Trained model artifacts
• endpoints.yml – Runtime endpoint config (Rasa server, action server)
• credentials.yml – Connector credentials (Slack, Facebook, etc.)
• config.yml – Pipeline & policy configurations
• Dockerfile & docker-compose.yml – Container definitions
• tests/ – Test stories
• dump/ – Static FAQ dumps, UI assets (HTML/CSS/JS)
README for additional insights
• It’s boilerplate; no architectural specifics. We infer from standard Rasa design.
Detailed checklist
• Separate Rasa NLU, Core, and Action Server as distinct services.
• Training pipeline: data/ + config.yml → train → models/.
• Channel connectors (HTTP, Slack, FB) into Rasa API.
• Action server calls out to third-party REST APIs.