How derxter/nummenu-bot works
Project type and purpose
• This is a Rasa-based chatbot (“NumMenu Bot”) that presents a numbered menu of deep-learning frameworks and returns descriptions.
• It uses Rasa NLU & Core for intent recognition and dialogue management, plus a custom action server (Python) to look up menu items.
File structure
Top-level files and folders:
• config.yml – Rasa NLU pipeline & policies
• domain.yml – intents, slots, responses & actions
• endpoints.yml – HTTP endpoints (action server)
• data/
– nlu.yml, rules.yml, stories.yml (training data)
• actions/
– actions.py & menus.py (custom actions & menu definitions)
• menu.yml – static menu definitions
• tests/ – test stories
• story_graph.dot – auto-generated story graph
Key patterns:
• Separation of NLU & Core (data/ + config.yml)
• Custom action server (REST endpoint defined in endpoints.yml)
• Static menu definitions loaded by actions
README for additional insights
• Describes training (rasa train), interactive learning, running two processes (rasa run actions + rasa shell).
• Mentions integration possibility (e.g., Twilio+WhatsApp) via Rasa connectors.
Main components and their interactions
a. Components
1) Client (shell or messaging platform like WhatsApp/Twilio)
2) Rasa Server (Core + NLU pipeline)
3) Custom Action Server (Python Flask/Starlette serving actions.py & menus.py)
4) Training Data & Model Store (config.yml, data/, domain.yml → models/)
5) External Services (e.g., Twilio API for WhatsApp)
b. Interactions
– User → Connector (shell/HTTP) → Rasa server
– Rasa server:
• Processes text through NLU pipeline → intent/entities
• Core uses stories & rules to decide next action
• If next action is a custom action → HTTP call to Action Server
– Action Server returns data (menu description) → Rasa returns response to user
– Optionally, Rasa logs events or publishes metrics
c. Architectural patterns
– Microservice-style separation (Rasa vs. Action Server)
– Declarative training data (NLU, Stories, Rules) → reproducible model
– External connector plugin approach for channels
d. Key technologies
– Python 3.x, Rasa Open Source
– YAML for configuration & training data
– Custom actions run over REST (Flask/Starlette)
– (Optional) Twilio/WhatsApp connector