FITNESS PLATFORM
Gamified Digital Health & AI-Powered Nutrition Intelligence
[Link] [Link] PostgreSQL TensorFlow MobileNetV2
Academic Pre-Production Document · 10 March 2026
PROJECT OVERVIEW
PROBLEM 1 PROBLEM 2
Motivation Decay Data Capture Friction
Conventional fitness apps rely on passive logging and fail
Manual nutrition and workout entry is tedious, leading to
to sustain long-term user engagement beyond initial
incomplete records and poor health decision support.
onboarding.
THE FORGE SOLUTION
FORGE unifies progression-based reward loops, XP ranking systems, and MobileNetV2-powered food
detection into a single cohesive environment — turning daily health tracking from a chore into a game.
AIMS &
O B J E C T I V ES
PROJECT AIM · Design and implement a scalable, modular fitness platform unifying behavioral reinforcement, performance
tracking, and AI-assisted nutrition intelligence into a cohesive digital health environment.
OBJ 1 Secure JWT-based auth & role-based access control OBJ 2 XP accumulation, task management & rank-tier engine
OBJ 3 Camera-based food detection via MobileNetV2 OBJ 4 Automated macro & calorie metadata enrichment
OBJ 5 Responsive, chart-rich dashboards for real-time insights OBJ 6 Strict module separation for independent scaling
OBJ 7 Capacitor mobile build path (Android & iOS)
SYSTEM ARCHITECTURE
CLIENT / PRESENTATION LAYER
[Link] 14 · React · TypeScript · Tailwind CSS · Capacitor (iOS / Android)
API GATEWAY & BUSINESS LOGIC
[Link] · Express · JWT Auth · Modular Route Handlers · Middleware Pipeline
DATA & INTELLIGENCE LAYER
PostgreSQL · ML Inference Service · MobileNetV2 · TensorFlow/Keras · Python RPC
Separation of Concerns Stateless REST API Orchestrated ML Mobile-First
FUNCTIONAL
MODULE BREAKDOWN
01 02 03
Auth & Profile Fitness Progression Food Intelligence
MobileNetV2 detection · Nutrition
JWT login · bcrypt hashing · RBAC · Profile XP engine · Task tracking · Rank tiers · Streak
enrichment · Macro computation · Meal
CRUD · Token refresh detection · Milestones
recs
04 05 06
Camera Capture Dashboard & Analytics Monitoring & Ops
MediaStream API · Capacitor Camera · XP trends · Rank history · Calorie balance · Health checks · node-cron scheduling ·
Image preprocessing · History Macro breakdowns · Charts Alerts · Operational logging
MACHINE LEARNING ARCHITECTURE
BASE MODEL
INFERENCE DATA FLOW
MobileNetV2
1 Image Capture
TR AIN IN G Camera / Upload
Transfer Learning · ImageNet Weights
Preprocess
2
RUNTIME 224×224 Resize + Normalize
TensorFlow / Keras (Python)
ML Inference
3
MobileNetV2 Classification
INPUT
224 × 224 px RGB · Normalised [0,1]
4 Class → Nutrition DB
Macro & kcal lookup
OUTPUT
Top-K Class Probability Vector (JSON)
5 API Response
Macros · Calories · JSON
TECHNOLOGY STACK
FRONTEND BACKEND DATABASE ML / AI
[Link] 14 (React 18) [Link] Runtime PostgreSQL Engine TensorFlow / Keras
TypeScript [Link] Framework pg (node-postgres) Driver Python 3.x Runtime
Tailwind CSS JWT Authentication Relational / Normalised MobileNetV2 Base Model
Recharts / [Link] bcrypt Password Hashing Versioned SQL Migrations ImageNet Transfer Learning
Capacitor (Android & iOS) node-cron Scheduler XP / Nutrition Logs HTTP RPC Inference Server
RESULTS & IMPACT
Higher Engagement Better Adherence
Progression loops — XP, rank tiers, streaks — replicate Consistent feedback and milestone recognition reduce
proven game mechanics, sustaining active platform use drop-off by reinforcing daily tracking habits through
well beyond initial onboarding. positive, reward-driven interactions.
Informed Nutrition Scalable Foundation
AI-assisted food detection removes manual logging Modular service separation allows independent
friction; nutrition enrichment transforms raw images into evolution of UI, business logic, and ML inference without
actionable macro & calorie insights. coupled releases or regression risk.
LIMITATIONS &
FUTURE WORK
CURRENT LIMITATIONS ✦ FUTURE ENHANCEMENTS
– CI/CD quality gate coverage needs broadening for robust Real-time pose estimation for form correction & rep counting
release confidence via on-device ML
– ML model governance & dataset versioning require Personalised adaptive recommendation engine tuned to
formalisation individual goal trajectories
– Pose estimation features scaffolded but not yet Enhanced analytics — retention telemetry, cohort analysis &
production-ready funnel dashboards
– Mobile app-store deployment pipeline needs dedicated Community features: leaderboards, challenge rooms & social
hardening phase progress sharing
CONCLUSION
Three-tier modular architecture ensures Progression-driven UX directly addresses the
scalability and long-term maintainability retention weakness of conventional apps
MobileNetV2-powered food detection
Technically mature pre-production platform
meaningfully reduces nutrition logging
ready for scalable deployment
friction
Gamified Fitness · Health Informatics · MobileNetV2 · TensorFlow · Full-Stack Web Platform · Behavior Reinforcement