FlowForm

Landing Page
Working out
Workout Summary
Building a workout
Building a rehab workout

🌱Inspiration

The idea for FlowForm was born from our own fitness journeys. When we first started working out, we often wished we had a complete resource to guide and tell us if our form was correct in order to avoid injuring ourselves. Personal trainers were too expensive, and while we learned through trial and error, online resources were overwhelming, mixing good information with bad.

We realized there was a gap in the market for an accessible, hands-on tool that makes fitness and physical therapy less intimidating. We built FlowForm to be the product we wished we had when we began our journey—a simple, conversational way to learn proper form and get fit safely.

🤔What it does

FlowForm is an AI personal trainer and physical therapist that uses Snap Spectacles to provide real-time form correction.

Generate Workouts: The web app helps users generate a personalized workout by selecting options like "rehab" or "workout," and then the difficulty, category, style, and duration.
AR-projection through the Snap Spectacles in real-time, during the workout, giving visual guidance simply by wearing the glasses.
AI-Powered: Our backend, powered by Cohere, uses a dataset of poses to build the custom workout routine.
Text-to-speech real-time coaching, instructing and encouraging users throughout the workout.
Feedback on form powered by Martian's LLM routing, making LLMs "discuss" about it together, ensuring more accurate and appropriate coaching.

🔨How we built it

Frontend: Typescript, Vite Backend: Python, Uvicorn, Cohere API, Martian API Database: MongoDB Real-Time Visual Guidance: Snap Spectacles, ElevenLabs, Cohere API Pose Scoring:

❗Challenges we ran into

We had a few big challenges, such as finding the right dataset and trained model(?). We also had trouble learning how to implement AR and connecting it to our web app, with the new and exciting Snap Spectacles. Martian's routing API was simple yet challenging; a complex issue of high credit cost, as well as workout generation time, held us back from developing even more exciting features.

🎖️ Accomplishments that we're proud of

From meeting at the Pearson Airport to submitting a project involving technologies we've never used, some being very new (such as the Snap Spectacles), or spontaneously deciding to attend Martian's workshop and adding LLM routing to our project, we couldn't be happier with how our project turned out.

🎓What we learned

Beyond the new technologies we delved into, we learned that with good communication and perseverance, a seemingly impossible task isn't only achievable, but the process can also be enjoyable.

🌟What's next for FlowForm - Real-time Form Correction with Snap Spectacles

In the future, we plan to expand our pose dataset to include a wider range of exercises and rehab routines. We would also love to implement machine learning to allow the system to learn from a user’s form over time and provide more personalized feedback. We could also explore adding a feature that tracks progress and provides a summary of workout performance. Finally, integrating text-to-speech with the Snap Spectacles would further increase the immersiveness of working out.