Medaid

A Safety‑First Multimodal AI Triage System for Rural Healthcare Education

Comment

Inspiration

The original Medaid project responded to a critical healthcare gap in rural and underserved communities: limited access to timely medical advice and clinical care. For the hackathon we pivot the product toward healthcare education — a preventative, scalable approach that empowers individuals and communities with knowledge to make safer health decisions, recognise warning signs, and know when to seek care.

Rationale: In many rural areas, low health literacy and linguistic barriers prevent people from understanding preventive care, interpreting basic test results, or following treatment instructions. Education reduces avoidable clinic visits, improves early detection, and enables long-term behaviour change.

What it does

Medaid (Education Edition) is an AI-assisted, multimodal health education platform designed for low-bandwidth, multilingual rural contexts. It provides short, actionable learning units that teach preventive care, symptom recognition, first-aid, medication adherence, and how to use local health services.

Key user flows:

Learn by Symptom: A user types or speaks a symptom (for example, "fever and cough" ) and receives an educational card: concise explanation, likely causes, simple home-care steps, preventive tips, and when to seek medical help.

Health Education Passport: A downloadable PDF that records completed lessons, quiz scores, personalized learning recommendations, and locally relevant resources (e.g., nearest PHC). This acts as both a learning record and a handoff for clinicians.

How we built it

Architecture & Stack

Frontend: Streamlit for rapid prototyping + a lightweight React/HTML fallback for production UI.

Backend: Modular Python services (FastAPI microservices) to serve content, manage users, and track progress.

Data store: MongoDB for user progress and content metadata, with a JSON fallback for offline caching.

Embeddings & Search: Vector search (FAISS) for semantic matching of user questions to lesson cards and verified guidance.

LLMs & NLU: Google Gemini (via LangChain) for generating lesson text, paraphrases, and explanations. Deterministic templates and a quality-control layer sanitize and standardize output.

OCR & Report Parsing: Optional module to extract simple values from lab reports and return educational explanations of key numbers (e.g., interpreting HbA1c in plain language).

Facility lookup: Pincode-based lookup for local health resources using Google Maps with OpenStreetMap fallback, but presented as educational links (how to reach, when to go).

Education-first Safety & QA Layer

Deterministic pre-checks detect emergencies ("severe chest pain", unconsciousness) and immediately return instructions to seek urgent care rather than educational content.

Content filters and fixed templates convert LLM outputs into structured lesson components (learning objective, 3–5 takeaways).

Key Features

Explain Like I'm 10 toggles for simpler explanations.

Health Education Passport: PDF summary of learned lessons, local referrals, and recommended next lessons.

Safety first layer to detect emergencies

100% accuracy on disease predictions

Challenges we addressed

Misinformation & Hallucination: Generative models can produce plausible-sounding but incorrect facts. We mitigate this by using templates, source attribution (WHO/MoHFW), and a deterministic finalizer that refuses to state uncertain medical facts as definitive.

Accurate disease prediction

Accomplishments

Safety‑first, learning‑ready architecture: The deterministic rule engine plus LLM pipeline lets us generate educational content while blocking unsafe or emergency outputs.

Nearby hospital recommendations

Disease prediction based on user history + current symptoms + report(if uploaded)

What we learned

Human validation is essential: Community health workers are invaluable for validating translations and cultural framing.

What's next

Pilot & Validation: Run a controlled pilot with 3–5 village health centers. Measure learning outcomes (pre/post quiz scores), behavior change (self‑reported clinic visits), and user engagement.

Curriculum Partnerships: Partner with public health NGOs and local health authorities to align lesson content with national public-health priorities (immunization, maternal health, nutrition).

Clinical Handoff Pathways: Build standard templates for handing over the Education Passport to clinicians, enabling continuity between education and care.

Built With

Share this project:

Updates