Maverick-Stan/maseru_health_ai

Maseru Health AI is a modular, safety-first healthcare support assistant built with Streamlit, rule-based escalation, and LLM orchestration for low-resource environments. # Maseru Health AI A modular AI healthcare support prototype exploring rule-based escalation, structured LLM workflows, and safety-aware response generation for low-resource environments. Live app: maseru-health-support-ai.st… ## Overview Maseru Health AI is a Streamlit-based healthcare support prototype for non-diagnostic conversations. It organizes a chatbot-style interface into a modular AI workflow with input parsing, rule-based safety checks, LLM-assisted guidance, structured response formatting, fallback handling, and deployment-conscious configuration. The project is intentionally conservative. It does not diagnose, treat, prescribe, provide emergency response, claim clinical validation, or present itself as production-ready. ## What the System Does - Accepts user messages through a Streamlit chat interface. - Extracts lightweight intent and support signals from user input. - Runs rule-based safety checks before the LLM is called. - Uses selected high-risk phrases and physical red-flag phrases to trigger escalation guidance. - Uses an existing demonstration TF-IDF + Logistic Regression risk-screening classifier as part of the safety layer. - Calls a single Google ADK + LiteLLM-backed LLM workflow when escalation is not required. - Generates structured internal responses and renders user-facing guidance in chat. - Returns a deterministic fallback response if the LLM is unavailable or fails. - Provides a developer/admin page for inspecting parser and safety metadata. ## What the System Does Not Do - It does not diagnose medical or mental health conditions. - It does not prescribe medication or treatment. - It does not replace clinicians, counselors, clinics, hospitals, or emergency services. - It does not provide emergency response capability. - It does not claim clinical validation. - It is not production-ready. - It does not implement retrieval-augmented generation. - It does not implement multi-agent orchestration. - It does not p …