Inspiration
We wanted to replace static, disconnected workout sheets and manual nutrition loggers with a unified, interactive AI personal trainer(specially for the introverts)
What it does
ZiddiFit is a closed-loop fitness app that combines real-time computer vision (MediaPipe) pose tracking with adaptive Gemini coaching and meal-to-workout plan updates.
How we built it
The stack leverages React 19, FastAPI, and Supabase, utilizing MediaPipe for local biomechanics tracking and Gemini Flash + Imagen for generative intelligence.
Challenges we ran into
We overcame camera joint-angle noise by writing custom EMA math smoothers and optimized real-time API latency for seamless voice feedback mid-workout.
Accomplishments that we're proud of
Can do real time pose correction and nutritional analysis.
What we learned
We mastered client-side edge compute (pose logic) paired with server-side LLM coordination to build a privacy-first, responsive user feedback loop.
Log in or sign up for Devpost to join the conversation.