How sirivalli-stack/mental-health-agent works
This is a research prototype for state-aware conversational mental-health support, not a clinical system. The main workflow is a user chat turn through the React interface and API, pre-generation safety, sentiment/emotion/risk analysis and session-state assembly, LLM response generation, then post-generation safety and presentation. Separate branches cover persistent memory and personalization, model training, and health/status reporting. Relationships reflect the excerpts, SOURCE INDEX references, or README; some wiring is documented rather than shown in sampled code.
Main entry points: engine.py User state engine · pipelines/chat.py Chat pipeline · api/chat.py Chat endpoint · text_normalise.py Text normalization
Components 24
Chat experience
- Chat interface
frontend/src/App.jsx - Frontend API
frontend/src/api.js
Conversation runtime
- Chat endpoint
backend/app/api/chat.py - Chat pipeline
backend/app/pipelines/chat.py - Request and state contracts
backend/app/models/schemas.py - Pre-generation safety
backend/app/safety/pre_generation.py - LLM generation
backend/app/services/llm_service.py - Post-generation safety
backend/app/safety/post_generation.py
State and memory
- User state engine
backend/app/state_engine/engine.py - Trajectory trends
backend/app/state_engine/trends.py - Session state store
backend/app/state_engine/store.py - Conversation memory
backend/app/memory/store.py - SQLite session database
backend/app/models/db.py - Profile extraction
backend/app/personalization/extractor.py
Inference and training
- Sentiment analysis
backend/app/services/sentiment_service.py - Emotion analysis
backend/app/services/emotion_service.py - Risk analysis
backend/app/services/risk_service.py - Classifier training
backend/app/training - Text normalization
backend/app/training/text_normalise.py
Safety and operations
- Risk rules
backend/app/safety/rules.py - Health endpoint
backend/app/api/health.py - Turn audit log
backend/app/pipelines/turn_log.py
Other components
- User
- Local LLM service
Connections 26
- User → Chat interface: sends message
frontend/src/App.jsx - Chat interface → Frontend API: submits chat
frontend/src/App.jsx - Frontend API → Chat endpoint: calls chat API
README.md - Chat endpoint → Chat pipeline: dispatches turn
backend/app/api/chat.py - Chat endpoint → Request and state contracts: validates request
backend/app/api/chat.py - Chat pipeline → Pre-generation safety: checks input
backend/app/pipelines/chat.py - Chat pipeline → User state engine: builds state
backend/app/pipelines/chat.py - User state engine → Sentiment analysis: infers sentiment
backend/app/state_engine/engine.py - User state engine → Emotion analysis: infers emotion
backend/app/state_engine/engine.py - User state engine → Risk analysis: infers risk
backend/app/state_engine/engine.py - Risk analysis → Risk rules: applies rules
backend/app/services/risk_service.py - User state engine → Trajectory trends: computes trends
backend/app/state_engine/engine.py - User state engine → Session state store: reads and updates
backend/app/state_engine/engine.py - User state engine → Conversation memory: recalls session
backend/app/state_engine/engine.py - User state engine → Profile extraction: extracts profile
backend/app/state_engine/engine.py - Conversation memory → SQLite session database: persists snapshots
backend/app/memory/store.py - Chat pipeline → LLM generation: generates reply
backend/app/pipelines/chat.py - LLM generation → Local LLM service: sends prompt
backend/app/services/llm_service.py - Chat pipeline → Post-generation safety: checks draft
backend/app/pipelines/chat.py - Chat pipeline → Turn audit log: records turn
backend/app/pipelines/chat.py - Frontend API → Health endpoint: checks service status
README.md - Classifier training → Text normalization: uses preprocessing
backend/app/training/risk_train.py - Sentiment analysis → Text normalization: uses preprocessing
backend/app/services/sentiment_service.py - Emotion analysis → Text normalization: uses preprocessing
backend/app/services/emotion_service.py - Risk analysis → Text normalization: uses preprocessing
backend/app/services/risk_service.py - Chat endpoint → Chat interface: returns response