-
Web Page
-
User Upload Page
-
Results(still working on backend)
Inspiration
In many road accidents, bystanders want to help but lack medical knowledge, and ambulances often take 5–15 minutes to arrive. This golden window can save lives — yet most people don’t know how to respond correctly. After witnessing such an incident personally, i was inspired to build FirstAidAI, a tool that guides people in real time with AI-powered first-aid instructions based on accident scene images.
What it does
FirstAidAI allows a user to take or upload an image of an accident scene. The system then: Uses Computer Vision and *Pose Estimation to detect injuries and the person’s position Identifies danger signs (e.g., head trauma, bleeding, unconsciousness) Provides step-by-step first-aid instructions: What to do / avoid Where to support the person Whether to give water How to move (if at all) This helps bystanders offer safe assistance until medical help arrives.
How we built it
We built the project using the following stack: Frontend: React.js / Flutter (mobile-first design) Backend: Python + Flask (REST API for predictions) Computer Vision: YOLOv8 (for injury detection), MediaPipe (for pose estimation) NLP: OpenAI LLM (for converting model output into readable instructions) Hosting: Firebase and Microsoft Azure (cloud + storage) Version Control: GitHub
Challenges we ran into
Finding open-source injury datasets for training models, Ensuring the instructions were medically accurate and non-harmful, Keeping the inference time low for real-time usage, Making the interface calm and usable during high-stress situations.
Accomplishments that we're proud of
created a working prototype capable of analyzing accident images and generating live first-aid steps, Integrated multiple ML models into a single pipeline, Made the system multilingual and mobile-friendly, Designed the system to be usable even by people with no medical or technical background.
What we learned
How to work with real-world healthcare data constraints, Designing AI systems for non-technical end users in critical situations, Building hybrid models combining CV, pose estimation, and LLMs, Importance of UI/UX in emergency tech.
What's next for FirstAidAI
🩺 Collaborate with medical professionals to fine-tune advice and expand injury database, 📶 Add offline mode for rural areas with no connectivity, 🧠 Train domain-specific LLM for faster and safer instruction generation, 🌐 Launch a pilot program with traffic police and public health NGOs.
