2025 Dream AI Hackathon Project
Build, analyze, and level up your fitness — with real-time AI coaching.
RoboLifter is an end-to-end AI-powered personal trainer that analyzes your squat and pushup form in real time using your device’s camera. It provides instant, accurate, and encouraging voice feedback using advanced pose detection, LLM-based coaching, and realistic text-to-speech.
It was built at the 2025 Dream AI Hackathon (Likelion US, Cambridge) — where 180+ innovators across 37+ teams built AI-driven solutions for real-world impact.
RoboLifter is designed not only as an AI fitness coach, but also as a fitness analytics platform capable of collecting and analyzing real-time workout performance data.
Planned analytics features include:
- Rep and workout history tracking
- Exercise performance trends over time
- Form accuracy scoring
- Fatigue and consistency analysis
- Workout session dashboards
- AI-generated progress insights
- Exportable workout datasets for further analysis
Future versions will integrate data visualization and analytics workflows using:
- Python Pandas
- SQL databases
- Tableau / Power BI dashboards
- Machine learning performance analysis
🔹 Get instant, personalized AI voice feedback
🔹 Track reps and progress automatically
🔹 Enjoy a modern, responsive web experience
TL;DR: Go from “start workout” to “AI-powered feedback” in <60 seconds.
# Clone the repository
git clone https://github.com/suriyaks0902/Ai-gymbro.git
cd Ai-gymbro
# Install backend dependencies
pip install -r requirements.txt
# Install frontend dependencies
cd app
npm install
npm run devSet your ElevenLabs API key for voice feedback:
export ELEVEN_API_KEY="your-elevenlabs-api-key"(Optional) Start Ollama for local LLM:
ollama run mistral- Start the backend:
uvicorn ai_server:app --reload
- Start the frontend:
cd app npm run dev - Open http://localhost:3000 in your browser.
- Allow camera access, select your exercise, and start moving!
- Full-stack AI fitness coaching: Real-time pose estimation, exercise detection, and voice feedback.
- Robust feedback API: Get actionable, encouraging tips for every rep.
- Modern UI: Responsive, mobile-friendly design.
- Multi-model support: Use any LLM (Ollama, OpenAI, etc.) and ElevenLabs for TTS.
- Accurate rep counting: Never miss a rep!
- Plug-and-play: Easy to set up and use.
| Exercise | Detection | Feedback | Rep Counting |
|---|---|---|---|
| Squats | ✅ | ✅ | ✅ |
| Pushups | ✅ | ✅ | ✅ |
Every rep or form error triggers a feedback event:
{
"eventType": "form_error" | "rep_complete",
"exercise": "squat" | "pushup",
"angles": {...},
"repCount": 3,
"formError": "Knees caving in",
"voice_id": "cgSgspJ2msm6clMCkdW9"
}- Personalized: Feedback is generated by an LLM and spoken using ElevenLabs TTS.
- Selectable voices: Choose male or female voice for your coach.
- Multilingual: Easily extend to other languages.
- Frontend:
/app/app/page.tsx— UI, camera, and pose logic. - Backend:
/ai_server.py— API, LLM, and TTS integration. - .env: Store your API keys securely.
2025 Dream AI Hackathon
Hosted by Likelion US, Cambridge
180+ participants | 37+ teams | 1 amazing weekend
- 📢 Feature requests? Open an Issue
- 🐛 Bug reports? Create a ticket
- 🛠️ Submit code? Fork + PR — we love clean commits!
Because everyone deserves a personal trainer — powered by AI, accessible anywhere, and always encouraging you to:
- reach your fitness goals.
- improve your form and stay motivated.
- unlock your full potential, one rep at a time.
- train smarter, not just harder.
- make fitness fun, interactive, and effective.