We will be undergoing planned maintenance on January 16th, 2026 at 1:00pm UTC. Please make sure to save your work.

GymIntel

GymIntel is an AI-powered personal trainer that analyzes your workout videos to provide instant form feedback, muscle activation tracking, and community benchmarking

Comment 1

Inspiration

We've all been there—you're at the gym, unsure if your squat depth is correct or if you're engaging the right muscles. Personal trainers are expensive, and looking in the mirror only gives you one angle. We wanted to build something that bridges the gap between expensive coaching and going it alone. We were inspired to create GymIntel, an intelligent system that "sees" your workout and gives you the kind of detailed, data-driven feedback that pro athletes get, accessible to everyone.

What it does

GymIntel is a comprehensive AI coaching platform. Users can upload videos of their sets, and our pipeline goes to work:

  1. Video Analysis: It identifies the exercise being performed and tracks specific joint movements.
  2. Form Feedback: It detects safety issues (like knee cave or rounded back) and scores your form execution.
  3. Muscle Mapping: It calculates "Muscle Balance" visualization, showing you which muscle groups you are over or under-training based on your activation data.
  4. AI Coaching: A built-in chatbot (powered by Gemini) acts as your personal coach, answering questions about your recent performance or giving tips for your next session.
  5. Benchmarking: It anonymously compares your stats (like range of motion and form consistency) against the community average, showing you where you stand.

How we built it

We built a full-stack application using Next.js for a responsive, modern frontend and FastAPI (Python) for the heavy lifting on the backend.

  • Video Processing Pipeline: We used TwelveLabs to index and "understand" the context of the workout videos, helping segment different exercises.
  • Computer Vision: We integrated YOLOv11 for pose estimation to track key points on the body and calculate joint angles for depth and form analysis.
  • Generative AI: We utilized Google Gemini to take the raw data (angles, reps, failures) and turn it into human-readable summaries and actionable advice.
  • Data: We used MongoDB to store user profiles, workout history, and the aggregated statistics used for our community comparison features.

Challenges we ran into

  • Video Pipeline Latency: Processing high-res video for pose estimation is computationally expensive. We had to optimize our backend to handle uploads efficiently.
  • Data Aggregation: Creating the "Public Comparison" feature required writing complex aggregation queries in MongoDB to calculate percentiles for form scores and range of motion without compromising user privacy.
  • Dependency Hell: We hit a tricky snag with Python's password hashing libraries (passlib vs bcrypt versions) that initially broke our authentication system, requiring some deep debugging.

Accomplishments that we're proud of

  • The "Muscle Map": We're really proud of the radar chart visualization that gives users an instant understanding of their training balance.
  • Seamless Integration: getting TwelveLabs, YOLO, and Gemini to "talk" to each other. The video upload triggers a chain reaction that results in a complete workout report without the user lifting a finger.
  • Actually "Watching": It's not just logging numbers; the system actually sees the depth of a squat and critiques it.

What we learned

  • The power of Multimodal AI: Combining vision models (YOLO/TwelveLabs) with language models (Gemini) creates a much richer experience than either could achieve alone.
  • We learned a lot about HLS streaming and how to deliver video content smoothly in a React application.
  • Designing for User Trust: When AI critiques your physical form, the feedback needs to be presented gently and constructively, or users might just quit.

What's next for GymIntel

  • Real-time Analysis: Moving from "upload and wait" to live feedback via webcam.
  • Social Features: Challenging friends to form competitions, sharing workouts, and leaderboards.
  • Wearable Integration: Syncing heart rate data to better estimate exertion levels.
  • Personal Records (PRs): Automatic tracking of max weight and reps per exercise.
  • Goal Setting: Weekly workout goals with intelligent progress tracking.
  • Workout Templates: Save and quickly reuse workout plans.
  • Progress Photos: Secure storage for before/after photos with AI body composition estimates.
  • Smart Notifications: Reminders to workout based on your schedule and streak celebrations.
  • Data Export: Downloadable CSV/PDF reports of training history for deeper analysis.

Built With

Share this project:

Updates