A Flask microservice that serves the machine learning models behind the
AI Fitness Coach platform's personalization features: workout plan
generation and nutrition/meal planning based on a user's profile.
This service is intentionally decoupled from the main backend API — it
only knows how to take structured input and return a plan, so the
models can be retrained, swapped, or scaled independently of the rest
of the system.
Tech Stack
Framework: Flask
ML/Data: scikit-learn, pandas, NumPy
Serving: Gunicorn (production WSGI server)
Deployment: Docker, configured for Hugging Face Spaces
What It Does
POST /fitness — Personalized Workout Plans
Given a user's training environment, experience level, goal, gender,
age, prior feedback, current working weight, and available equipment,
returns a 30-day workout plan.
Under the hood this combines two models:
A trained plan classifier picks a body-part split template based
on the user's level, goal, and gender (e.g. which muscle groups to
train on which days across a rotation).
A KMeans clustering model, fit on an exercise dataset grouped by
level/goal/body-part, is used to select a specific exercise for each
day — filtering out exercises that need unavailable equipment when
the user is training at home.
On top of the model output, a set of rule-based adjustment functions
personalizes each exercise's working weight, rep count, and duration
using age, gender, experience level, and whether the user gave positive
or negative feedback on their last session — so two users with the same
cluster assignment still get different loads.
POST /nutrition — Weekly Meal Plans
Given a daily calorie target, splits it across breakfast (30%), lunch
(40%), dinner (30%), and a snack, builds a nutrient profile for each
meal slot (calories, fat, protein, carbs, fiber, sodium, etc.), and uses
a trained model to match each slot against a real meals dataset —
returning a 7-day meal plan built from actual recipes rather than
generic macros.
1# Install dependencies2pip install -r requirements.txt
34# Run the server (dev)5python server.py
67# Run with Gunicorn (production-style)8gunicorn -b 0.0.0.0:7860 server:app
The service listens on port 7860 by default (configurable via the
PORT env variable) and exposes a health check at /.
Docker
bash
1docker build -t ai-fitness-coach-model-server .2docker run -p 7860:7860 ai-fitness-coach-model-server
