FitCoach AI
A chat-based fitness coach that builds a real plan around your situation.
Overview
FitCoach AI is a personal trainer you talk to. Rather than handing you the same template as everyone else, it asks about your goals, your gym, and what's getting in your way, then builds a workout and nutrition plan that fits. It also has a light XP and achievements system, because the real problem with fitness apps isn't the plan, it's that people quit in week one.
The problem
Most fitness apps give everyone the same plan, and a real coach costs more than most people are willing to pay every month. The advice that actually works is the advice that fits your life: the time you have, the equipment you can get to, the knee that plays up.
So the gap was a coach that feels genuinely personal and keeps you showing up, without charging like a human trainer.
The solution
I built a coach that gets to know your situation before it tells you anything, then gives you a plan you can actually follow. The progression system, with levels and unlockable achievements for consistency, is there to turn 'I'll start Monday' into a streak you don't want to break.
It's one focused chat with a progress panel next to it, so the plan and the motivation to stick to it live in the same place.
Architecture
The chat frontend streams the conversation to a server route that runs the coaching model. Progress and achievements are tracked on the client against clear milestones.
- Asks about your situation before it gives advice
- Model calls run on the server, so keys and logic stay private
- Progression system built to keep people coming back
- Coaching and motivation on one screen
Screens & diagrams
shotKey features
Gets to know you first
Asks about goals, equipment, and limits before it recommends anything.
A real plan
Gives you a workout and nutrition plan shaped by your situation, not a template.
XP and levels
Progression and milestones turn a good intention into a streak.
Achievements
Badges you unlock for staying consistent, which keeps the momentum going.
Implementation highlights
- Shaped the conversation so the model gathers enough context to give advice that's actually worth following.
- Spent real effort on the engagement layer, because keeping people past week one matters more than the plan itself.
- Kept model calls on the server and streamed the replies so the chat feels quick.
- Wrapped the model call in timeouts and clean error states so the chat never just hangs.
Challenges solved
Personal, not generic
The whole value is in the fit. Front-loading the questions is what made the plans feel like coaching instead of a lookup table.
Getting people to stick
A good plan is useless if you stop after four days, so the progression system ended up being as important as the AI.
Technologies
Future improvements
- Save plans and log the sessions you actually did
- Pull in wearable data to adjust the plan automatically
- Shareable streaks and challenges with friends