Menu
Healthcare · Completed
Adaptive Fitness Coach
One platform connects adaptive planning, live exercise feedback, nutrition, health data, analytics, and AI coaching.
- Duration
- 3–4 month build
- Team
- Project Lead (Me) + Mobile, Backend & QA Teams
Overview
Overview
Personal fitness plans often begin with generic recommendations that cannot respond to someone’s goals, physical profile, or actual workout consistency. The company wanted a mobile experience that could personalize training and nutrition while keeping users engaged between sessions.
I led the project from client communication through technical delivery. I translated product feedback into specifications, coordinated the mobile, backend, and QA team, presented weekly demonstrations, and developed the Python-based exercise tracking system.
The completed platform combines adaptive workout plans, meal planning, camera-based posture and repetition detection, fitness analytics, scheduled engagement, fasting tools, health integrations, and a domain-constrained AI coach.
Confidentiality note: Client and company names have been changed or omitted for confidentiality. The work, my role, and project scope are real.
Approach
Challenges & Solutions
A personalized starting point
Onboarding needed to turn personal goals, preferences, and body-scan information into a practical fitness plan.
Goal-based planning
Combined onboarding preferences, fitness goals, and body-scan outputs to support personalized workout recommendations.
Real-time exercise understanding
The mobile camera needed to recognize body movement, evaluate exercise posture, and count repetitions during active sessions.
Live movement analysis
Developed a Python program using MediaPipe and OpenCV to track body landmarks, calculate movement angles, detect posture, and count exercise repetitions.
Plans that adapt
The next week’s recommendations needed to consider workout completion, frequency, and previous performance instead of repeating a static schedule.
Adaptive fitness and nutrition
Connected workout adherence with weekly plan adjustments while supporting personalized meals, fasting, calorie estimation, and food scanning.
Useful but controlled AI
AI features needed to remain focused on the fitness domain while balancing practical user access with model-usage costs.
Domain-constrained AI
Integrated Google Gemini models for scanning and coaching features, restricted interactions to the application’s fitness logic, and applied practical usage limits.
Technical Detail
How It Was Built
The mobile experience was built with Compose Multiplatform. NestJS and PostgreSQL supported the application APIs and data, while Google Gemini models powered the body, food, planning, and coaching experiences.
Real-time exercise detection ran through a dedicated Python program using MediaPipe and OpenCV. It interpreted body landmarks and joint angles to identify movement, evaluate posture against expected exercise behavior, and count repetitions.
As project lead, I managed the client relationship, weekly demonstrations, feedback translation, technical specifications, implementation strategy, team coordination, and delivery oversight across mobile, backend, AI, and QA work.
Stack
Built With
- Compose Multiplatform
- NestJS
- PostgreSQL
- Python
- MediaPipe
- OpenCV
- Google Gemini API
Client Feedback Summary
Client Feedback
The AI features made the platform feel like a personal coach for someone exercising alone, not just another fitness app. Weekly workout planning, repetition counting, posture guidance, and the guardrails around those AI experiences were what surprised and impressed us most.
Client name withheld
Client identity is confidential. This summary reflects verbal feedback shared with Muhammad and the project team.
Facing Similar Challenges?
Let's Talk About What You Need Next.
Share what you are building or improving, and I'll help you identify a practical next step.
Discussion
Comments
No comments yet. Be the first to share your thoughts.