Kinetix

Kinetix is your AI personal trainer. Using your webcam, it analyzes your exercise form in real-time, providing instant visual and spoken feedback to ensure perfect technique and prevent injury.

Comment

Project Overview

Kinetix is a new application built using the Gemini 3 API that provides real-time exercise form feedback using computer vision and generative AI. The goal is to reduce injury risk caused by poor exercise form by providing immediate coaching without requiring a personal trainer.

Functionality

Kinetix uses a webcam feed to analyze user movements during exercise.

  • Frame-by-frame analysis of posture and joint alignment
  • Classification of movements as safe, suboptimal, or unsafe
  • Real-time visual and spoken feedback

Dual-Persona AI Design

Gemini 3 operates under two enforced roles:

  • Analyst (internal): Performs technical reasoning about biomechanics and movement errors
  • Coach (external): Converts the analysis into short, clear instructions suitable for users during exercise

The internal analysis is not exposed to the user.

Visual Feedback

  • Green screen border indicates correct form
  • Yellow indicates minor corrections
  • Red indicates potentially unsafe movement
  • Skeleton overlay highlights posture and joint positioning

Session Summary and Recommendations

After each workout session, Gemini 3 generates:

  • Summary of correct repetitions and detected form issues
  • Identification of recurring problem areas
  • Recommended corrective exercises based on observed weaknesses

Technical Implementation

Frontend

  • Next.js and React with TypeScript
  • Tailwind CSS and ShadCN UI
  • Real-time overlays designed to minimize distraction during movement

Backend

  • Python backend using FastAPI
  • WebSocket communication for low-latency updates

Use of Gemini 3 API

The application is centered around Gemini 3.

  • Processes visual input from video frames to reason about pose and movement
  • Structured system prompt enforces Analyst and Coach roles
  • Generates real-time coaching text
  • Produces post-session summaries and exercise recommendations
  • Used for text-to-speech feedback during workouts

Challenges

  • Designing prompts that balance technical correctness with simple user-facing feedback
  • Maintaining low latency for real-time analysis
  • Designing a UI that communicates errors without distracting the user

Outcomes

  • Working real-time exercise coaching system using a standard webcam
  • Structured Gemini 3 outputs suitable for direct UI integration
  • End-to-end feedback loop: analysis, guidance, summary, and correction

Future Work

  • Support for additional exercises and movement types
  • Long-term session storage for progress tracking
  • Guided workouts generated by Gemini 3

Submission Notes

  • New application built using the Gemini 3 API
  • All team members added to the submission
  • Application code included in the submitted zipped folder
  • Public demo video shows the system operating end-to-end

Built With

Share this project:

Updates