Coachify

Coachify — Your body. Your ceiling. Your coach.

Comment

Inspiration

We kept seeing the same problem, people following generic fitness plans built for someone else's body. A 5'8" narrow-framed ectomorph and a 6'1" wide-framed mesomorph don't need the same program. We wanted to build something that starts with your biology, not a template.

What it does

Coachify analyzes your body structure using computer vision, classifies your genetic archetype across 9 body profiles, and with the help of the ML and Agentic AI generates a complete transformation blueprint, realistic peak physique, personalized macros, optimized workout split, and a consistency-adjusted timeline is built. An 80% consistent user gets an aggressive plan. A 40% consistent user gets a sustainable one. Same body. Different coach.

How we built it

Computer Vision - MediaPipe pose estimation extracts skeletal landmarks from front and back photos, computing shoulder-hip ratio, frame type, and body composition ML Model - scikit-learn classifier maps inputs to one of 9 genetic archetypes (3 frames × 3 composition states) Agentic AI - Claude API with tool calling dynamically selects workout, diet, and timeline engines based on consistency score Backend - FastAPI + SQLAlchemy + SQLite handling auth, user profiles, and transformation plans Frontend - Next.js + Tailwind + shadcn/ui with a 6-step intake form, animated analysis screen, and 3-tab dashboard

Challenges we ran into

CV measurement accuracy - Extracting real-world centimeter estimates from 2D photos without depth data required careful calibration using height as a reference scalar Consistency scoring - Translating lifestyle inputs (sleep, stress, skip frequency) into a single 0–1 score that meaningfully changes the AI's recommendations took several iterations 24 hours - Three engineers, one night, zero sleep. Scope management was the hardest engineering decision we made

Accomplishments that we're proud of

Built a full CV → ML → Agent → Dashboard pipeline from scratch in under 24 hours The consistency-adaptive coaching logic - same body data, meaningfully different plans based on real behavior A UI that looks and feels like a real product, not a hackathon project Successfully integrating 5 different technologies into one coherent user flow

What we learned

Computer vision on unconstrained photos is genuinely hard - lighting, pose variation, and clothing all affect landmark detection significantly Agentic AI with tool calling is remarkably powerful for adaptive planning when the tools are well-defined Scope ruthlessly. The features that didn't make it in 24 hours taught us as much as the ones that did

What's next for Coachify

Progress tracking - monthly photo re-analysis to measure actual vs predicted adaptation rates and recalibrate the plan AI-generated physique previews - Gemini-powered visualization of your body at 3, 6, and 12 months Wearable integration - Apple Health and Whoop data to automatically update consistency scores in real time Expanded genetic modeling - moving beyond 9 archetypes to continuous physiological modeling using a larger training dataset

Built With

Share this project:

Updates