First-Aid Buddy: An Intelligent Emergency Assistant Integrating Large Language Models and RAG

First-Aid Buddy is an intelligent emergency assistant built with Python and Streamlit. It combines large language models (LLM), retrieval-augmented generation (RAG), and real-time external APIs to provide step-by-step guidance for users facing medical emergencies. Its core goal is to bridge the gap between ordinary users and professional first-aid knowledge, helping people respond correctly in critical moments.

In emergency medical situations, every second counts. However, most people feel helpless when facing sudden medical events and don't know how to respond properly. First-Aid Buddy was created to solve this pain point. Developed by tuvakel, it uses Python and Streamlit, integrating cutting-edge AI technologies like LLM, RAG, and real-time external APIs to understand user descriptions and provide accurate, actionable guidance based on authoritative first-aid knowledge.

LLM Integration

LLM (e.g., GPT series, Claude) plays key roles: symptom parsing, risk assessment, step-by-step guidance, and calming communication.

RAG System

To avoid LLM hallucinations, RAG retrieves relevant fragments from an authoritative first-aid knowledge base first, then uses them as context for LLM to generate answers. Benefits: improved accuracy, traceability, easy knowledge updates, and domain customization.

Real-Time External APIs

Integrated APIs include: location services (nearby medical institutions), weather info (for heatstroke/frostbite), medical resource databases (hospitals/pharmacies), and emergency service interfaces (contacting emergency centers).

Streamlit Frontend

Streamlit was chosen for its fast development (pure Python), elegant UI, real-time interaction, and easy deployment, allowing developers to focus on core functions.

Typical Use Cases

Applicable to family accidents (burns, cuts), outdoor emergencies (heatstroke, sprains), chronic disease management (asthma, hypoglycemia), and mental first aid (panic attacks).

Interaction Flow

1. User describes the situation.
2. System clarifies key details.
3. Preliminary assessment (immediate medical attention or self-handling).
4. Step-by-step guidance.
5. Follow-up suggestions.
6. Resource recommendations (nearby hospitals if needed).

Technical Highlights

• Multimodal input: text, voice, image upload (for symptom analysis).
• Context memory: maintains dialogue history to avoid repeated descriptions.

Safety Mechanisms

• Emergency recognition: prioritizes calling emergency services for critical cases.
• Confidence threshold: advises professional help for uncertain situations.
• Disclaimer: clarifies it's an auxiliary tool, not a substitute for professional medical advice.
• Knowledge boundary: clearly states its capabilities and limitations.

• First-Aid Knowledge Popularization: Reduces avoidable deaths due to lack of first-aid knowledge.
• Medical Resource Optimization: Helps users judge whether to seek immediate medical help, reducing unnecessary ER visits.
• Special Group Care: Provides extra safety for the elderly living alone, chronic patients, and new parents.

Future Directions

1. Multilingual support.
2. Offline mode for basic first-aid queries.
3. Integration with wearable devices.
4. AR-assisted guidance.
5. Community function for user mutual help.

Conclusion

First-Aid Buddy demonstrates how LLM and RAG can add value in medical emergencies—acting as a bridge between users and professional guidance, not replacing medical staff. It also serves as a technical reference for integrating LLM, RAG, Streamlit, and APIs in other fields.

Project address: https://github.com/tuvakel/First-Aid-Buddy