BeFit

• 

• 

Inspiration

I was inspired to build BeFit because I saw how many people struggle to stay consistent with their workouts. I wanted to make a fitness app that doesn’t just show static workout plans, but creates personalized workouts and helps users improve their form using real-time AI feedback. I also love AI tools and wanted to bring them into fitness in a meaningful way.

What it does

BeFit is an intelligent AI-powered workout assistant that combines Agentic AI, RAG-enhanced knowledge, and MediaPipe pose detection to create personalized workout routines and provide real-time form correction. it is designed for fitness enthusiasts at every level.

How we built it

I built BeFit with these main ideas:

SvelteKit & Tailwind CSS for a clean, responsive frontend UI.

AI tools & LLMs to generate workout routines based on goals, experience, and equipment.

Pose detection to analyze form and give real-time feedback during workouts.

Database & storage so workout routines and exercise configs are saved and reusable.

Users can tell BeFit what their goals are, and the AI will give them custom workout plans tailored just for them, not the same generic plans everyone sees in other apps. The app also analyzes movement through the webcam and gives real-time feedback to help improve exercise form.

Challenges we ran into

There were a few big challenges along the way:

Integrating AI with real-time feedback: Getting the AI to generate accurate workouts and usable pose configs took experimentation.

Learning new tools: I had to learn how to work with pose detection tech and connect it to AI responses.

Balancing features with usability: I wanted the app to be powerful but also simple to use. That meant spending extra time refining UI and interactions. But overcoming these challenges taught me a lot and made the project much stronger.

What we learned

While building this project, I learned a lot about:

AI-powered apps - how to use Large Language Models (LLMs) to generate workouts and respond to users.

Real-time pose analysis - integrating pose detection (using MediaPipe) to give users feedback on their exercise form.

Full-stack development - connecting frontend UI with backend services, databases, and AI tools.

User-centric design - making features that feel helpful and easy to use for anyone at any fitness level.

What's next for BeFit

Next, I want to make BeFit even smarter and more helpful.

First, I plan to improve the AI workout generation so it can adapt over time. Instead of just creating a workout once, BeFit will track progress and adjust routines based on performance, consistency, and user feedback.

I also want to make the pose detection more accurate and expand it to support more exercises. This will help users get better real-time feedback and reduce the risk of injury.

Another big step is adding:

📊 Progress tracking dashboard (charts, streaks, strength improvements)

🎯 Goal-based plans (fat loss, muscle gain, flexibility, home workouts, etc.)

📱 Mobile optimization or a dedicated mobile app

🧑🏽‍🤝‍🧑🏽 Community features like challenges and leaderboards

Long term, I want BeFit to feel like a real AI fitness coach, not just an app that gives instructions, but one that learns from you and grows with you. The goal is simple: make fitness more personal, more intelligent, and easier to stay consistent with.

Built With

• openai
• opik
• postgresql
• prisma
• qdrant
• typescript
• vercelaisdk