StudentHealth360: An Intelligent Health Risk Assessment Agent System Based on LangGraph

StudentHealth360 is an intelligent health risk assessment agent system based on LangGraph, which upgrades traditional machine learning risk prediction to an Agentic AI workflow. It uses LangGraph to orchestrate multi-node reasoning processes, combines RAG to retrieve medical guidelines, and generates structured health reports containing risk summaries, preventive recommendations, and source citations. The tech stack includes LangGraph, Groq API (Llama3), ChromaDB, Streamlit, etc. It is suitable for scenarios such as university health centers and health education, with both practicality and ethical considerations.

Traditional predictive medical systems only generate numerical risk scores, lacking intelligent interpretation and continuous support, which reduces their practical application value. University students' health is affected by multiple factors such as physiology, behavior, and psychology; simple risk level prediction is insufficient to provide meaningful suggestions and personalized guidance. StudentHealth360 aims to address this pain point by transforming static ML prediction systems into dynamic reasoning-driven AI assistants, combining LangGraph orchestration and RAG knowledge retrieval.

System Architecture

1. Data Input and ML Prediction: Users input information via Streamlit, call a logistic regression model to return risk scores and top 3 contributing features as the Agent's input layer.
2. LangGraph Agent Workflow: After loading risk scores and features into AgentState, it executes risk analysis (Groq API reasoning), RAG retrieval (ChromaDB semantic matching), and hallucination prevention (prompt engineering constraints) in parallel.
3. Report Generation: Integrate outputs to generate structured reports (risk summary, preventive recommendations, cited sources, disclaimers), including error handling nodes to ensure graceful degradation.

Tech Stack Selection

Layer	Tech Choice	Reason for Choice
Agent Framework	LangGraph	Supports complex workflows and state management
LLM	Groq API (Llama3)	Sufficient free quota and fast response
Vector Database	ChromaDB	Lightweight and easy to deploy locally
Embedding Model	all-MiniLM-L6-v2	Small size and good semantic performance
ML Model	Scikit-learn	Mature, stable, and serializable
UI Framework	Streamlit	Quickly build data applications
Deployment Platform	Hugging Face Spaces	Free hosting and easy to share

RAG Implementation Details

• Knowledge Base Construction: Medical guideline texts are embedded via sentence-transformers and persisted to ChromaDB vector storage.
• Retrieval Strategy: Queries are constructed based on predicted risk categories to extract the most relevant guideline fragments, with stronger focus.
• Prompt Engineering: Constrains LLM to only use retrieved results, annotate uncertain content, prohibit making up recommendations, and each recommendation must be labeled with sources.

Ethical AI Considerations

• Hallucination Prevention: Outputs are strictly based on guidelines; no claims without source attribution are allowed.
• Disclaimer: Both reports and UI include disclaimers that AI recommendations do not replace professional medical advice.
• Data Privacy: Patient data is processed locally during sessions and no records are stored.
• Bias Awareness: Non-predictive demographic variables are removed during ML model training to reduce bias.

StudentHealth360适用于以下场景：

1. University Health Centers: Preliminary risk assessment and lifestyle recommendations, identifying high-risk individuals.
2. Health Education: Improve students' health awareness and provide actionable preventive suggestions.
3. Research Prototype: Verify the application value of Agentic AI in the medical field.
4. Teaching Example: Serve as a teaching project for ML, NLP, and AI Agent courses, demonstrating the evolution from traditional ML to Agent architecture.

Current Limitations

• Relies on free API quotas; large-scale deployment requires paid plans.
• Limited coverage of medical guidelines; the knowledge base needs continuous expansion.
• Only targets university students; generalization to other groups requires retraining.

Future Improvement Directions

• Integrate more medical data sources (papers, clinical guideline databases).
• Support multi-language report generation.
• Add conversational interaction to allow users to ask follow-up questions for clarification.
• Implement long-term health tracking and trend analysis.

StudentHealth360 successfully upgrades traditional ML systems to Agentic AI applications. Through LangGraph orchestration, RAG knowledge enhancement, and strict hallucination prevention, it achieves the transformation from static prediction to a dynamic intelligent assistant. Its practical tech stack selection, clear code structure, and sufficient ethical considerations make it an excellent reference project for learning and practicing AI Agent development.