Fittie — Inclusive AI-Powered Adaptive Fitness Coach

Fittie uses AI to adapt fitness to your energy. Feeling drained? Get a Zen flow. Hyped? Power HIIT. Build streaks with a cute companion, not burnout.

Comment

🐻 Fittie — Inclusive AI-Powered Adaptive Fitness Coach

💡 Inspiration

The inspiration for Fittie stems from a glaring gap I saw in the fitness industry: the lack of empathy for the "daily ebb and flow" of human energy. Traditional fitness apps rely on rigid, static templates that ignore how life actually happens, stressful workdays, poor sleep, or chronic health fluctuations.

I was deeply moved by Spoon Theory, a conceptual model used by chronic illness communities to manage limited energy reserves. I wanted to build a coach that speaks that language. By combining this empathetic core with a bold Neo-Brutalist aesthetic and a friendly kawaii polar bear mascot, I developed Fittie as a Beta proof of concept to transform fitness from a chore into a highly personalized, accessible dialogue where no one is truly left behind.

🚀 What it does

Fittie is an all-in-one AI fitness companion designed to prove that inclusivity and high-performance tech can coexist. It doesn't just give a workout; it builds one specifically for the "me" that showed up today.

  • Real-Time Workout Generation: I ensured that no two workouts are ever the same. Gemini 3 Flash generates unique routines based on energy levels (0-100% or 1-5 "spoons").
  • Multimodal Gym Scanning: I built a system where users can photograph their environment, and Fittie identifies available equipment to build a workout around what's actually there.
  • Inclusive-First Coaching: I integrated specialized logic for wheelchair users and 11 chronic conditions (including POTS, MS, and Fibromyalgia) to ensure safety is never an afterthought.
  • Morphic UI: I designed the interface to physically transform—mascot personality and voice tone included—shifting between Power, Zen, and Desk modes to match the user's biometric vibe.
  • Hands-Free Voice Coaching: I used ElevenLabs narration and voice commands to allow for screen-free, focused sessions.

🛠️ How I built it

I developed Fittie as a cross-platform application using Flutter 3.6.0. My architecture leverages a "Triple Gemini" (with the Gemini 3 Flash model) strategy to handle complex reasoning:

  1. Workout Generator (JSON Mode): Processes 800+ exercises and uses MET-based math to calculate precise calorie burn. Uses exerciseDB, the gallery for visuals are still lacking for those with mobility issues.
  2. Vision Analyzer (Multimodal): Interprets gym photos to return structured equipment lists and space assessments.
  3. Chat Coach (Long Context): Replays the last 40 messages and the last 50 workout logs into each session to provide history-aware, progressive advice.

The backend is powered by Firebase (Auth, Firestore) for persistent chat memory. I handled natural voice synthesis via ElevenLabs, while global state management is orchestrated by Provider.

Scientific Accuracy

To ensure Fittie provides real value even in its Beta stage, I calculate energy expenditure using the Metabolic Equivalent of Task (MET) formula:

$$\text{Calories Burned} = \left( \frac{\text{MET} \times 3.5 \times \text{weight in kg}}{200} \right) \times \text{duration in mins}$$

🧠 Challenges I ran into

One of the primary technical hurdles I faced was Multi-Tier Exercise Matching. My AI might suggest an "Explosive Floor Press," but my database only contains a "Floor Press." I engineered a complex matching system using synonym maps and token intersection scoring to ensure users always have a visual GIF guide.

Additionally, maintaining Cross-Session Memory required a robust "Context Replay" system. I had to carefully curate the data sent to the Gemini API to prevent hitting token limits while ensuring "Fittie the Bear" remembered critical user details, like a knee injury from months prior.

🏆 Accomplishments that I'm proud of

  • Accessible Engineering: I am incredibly proud of the Spoonie Scale and Wheelchair Mode. Seeing the AI intelligently swap a standing squat for a seated variation based on a single profile toggle was a major win for me.
  • The Custom Mascot: Using CustomPaint, I built an animated polar bear that breathes, blinks, and talks in real-time without the performance overhead of traditional video assets.
  • Gemini 3 Integration: I successfully utilized 5 distinct Gemini capabilities (Long Context, Vision, JSON Mode, Chat, and Reasoning) to create a cohesive, intelligent experience.

📖 What I learned

This project taught me the true power of Context-Aware AI. I discovered that LLMs are powerful reasoning engines that can handle complex physical constraints better than any hard-coded algorithm. I also deepened my understanding of Neo-Brutalist Design, learning how to balance high-contrast aesthetics with a user-friendly, accessible interface.

🗺️ What's next for Fittie

As a proof of concept, the journey for Fittie is just beginning. My roadmap for moving beyond this Beta phase includes:

  • Wearable Integration: Syncing real-time heart rate data to the "Morphic Engine" for automatic mode switching.
  • Social Challenges: Allowing users to share their AI-generated "Flows" with friends via the Community Blog.
  • Advanced Vision: Moving from static photos to real-time video form correction during workouts. And expanding visualizations for those with mobility issues and other health issues.
  • Global Expansion: Localizing Fittie's coaching and mascot personality for different cultures to ensure no one, anywhere, is left behind.

Built With

Share this project:

Updates