FitAI
An AI-first fitness app that generates personalised workout plans, imports existing ChatGPT plans, and coaches users through every rep with video demonstrations and voice guidance.
fitai-hero-1.png
375 x 812
fitai-hero-2.png
375 x 812
fitai-hero-3.png
375 x 812
The Problem
Millions of people use ChatGPT to generate workout plans. They get a well-structured plan — exercises, sets, reps, rest periods — and then have absolutely no way to follow it. The plan lives in a chat window. There are no exercise demonstrations, no tracking, no progression, no accountability.
Existing fitness apps take the opposite approach — rigid pre-built programs that don't adapt to the user's goals, equipment, injuries, or schedule. You either follow their plan exactly or you're on your own.
The gap is clear: people want personalised plans (which AI can generate) paired with guided execution (which requires a proper app experience). No product bridges this gap today.
FitAI was built to own this intersection — AI-generated plans you can actually follow, with video coaching, voice cues, and smart progression. Import the plan you already have, or let FitAI build one from scratch.
The Solution
A PWA-first app (with parallel native builds via Expo) built on Supabase, powered by Claude for plan generation, ElevenLabs for voice coaching, and a curated exercise content library with safety-first content policies.
AI Plan Generation
Users answer a 7-screen onboarding questionnaire — goal, fitness level, available equipment, injuries, time per session, days per week, dietary preferences. Claude Sonnet generates a personalised multi-week workout plan in under 8 seconds. Plans are cached by archetype to keep LLM costs below ₹5/user/month.
fitai-plan-gen.png
375 x 812 (mobile)
Workout Player
A full workout playback experience — exercise sequence with video demonstrations or annotated diagrams, voice coaching via ElevenLabs narration, rep/set tracking, rest timers, and session completion logging. Compound barbell lifts always show diagrams (never AI-generated video) as a safety measure enforced at the database level.
fitai-player.png
375 x 812 (mobile)
Plan Import from ChatGPT
The wedge feature. Users paste their existing ChatGPT-generated workout plans as text, and Claude extracts structured exercise data with confidence scoring. Plans with 95%+ confidence are automatically converted into trackable FitAI plans. This captures users who already have a plan but no way to follow it.
fitai-import.png
375 x 812 (mobile)
Exercise Library & Content Pipeline
A curated library of exercises with AI-generated video clips for safe movements (yoga, calisthenics, isolation), annotated diagrams for compound lifts, and ElevenLabs voice narration for form cues. A content_type field is permanently locked per exercise at the database level — compound lifts can never be switched to AI video.
fitai-library.png
375 x 812 (mobile)
Key Decisions
Content Safety by Default
A database CHECK constraint permanently locks the content_type field on exercises. Compound barbell lifts (squats, deadlifts, bench press) can never display AI-generated video — only verified diagrams. This is an injury liability decision baked into the schema, not application logic that someone can accidentally change.
LLM Cost Ceiling
Every user has a hard ₹5/month LLM spending cap enforced via a database trigger on the ai_generations table. When exceeded, plan generation falls back to cached archetypes. This makes unit economics predictable from day one — not something to figure out post-launch.
PWA First, Native Parallel
The PWA ships first for fastest distribution (just a link, no app store). Expo React Native builds in parallel, sharing 70% of code via packages/shared. Both share the same Supabase backend. Friends test on PWA; TestFlight and Play Store internal track follow 3 weeks later.
Tech Stack
Frontend
Backend
AI & Media
Monorepo
Analytics
Deploy
Built-in Engagement
Streak System
Current streak, longest streak, freeze protection. Visible on the home screen, updated after every completed session.
Daily Check-ins
Mood, energy, and adherence tracking. Feeds back into plan adjustment — if energy is consistently low on Fridays, the AI adjusts intensity.
Gems & Rewards
Earn gems for completing workouts, maintaining streaks, and hitting milestones. Redeemable for premium content access.
Content Tabs
Yoga library (10 guided sessions), meditation tracks with a content filter, and a locked-content teaser mechanic that drives engagement and upgrades.
Admin Panel
User lookup, manual subscription grants (90-day free for beta testers), gem grants, content takedown with 4-hour SLA, activity audit log, and push notification triggers.
In-app Feedback
Built-in feedback widget captures structured input from beta testers — category, description, and device info — feeding directly into the development backlog.
Results
Building an AI-powered product?
From LLM integration and cost management to content pipelines and engagement systems — I build AI-first products with production-grade infrastructure. Let's talk.