How I Built an AI Gym Coach for People Who Train Alone

This new year I made a decision I had been putting off for a long time. I was going to take my health seriously. I signed up for a gym membership, bought proper shoes, watched YouTube videos on form, and showed up.

And then I stood in the corner not really knowing if I was doing anything right.

I am not someone who finds it easy to ask for help in public. The gym felt like everyone else already knew what they were doing and I was the only one guessing. I could not tell if my squats were deep enough, if my back was rounding on deadlifts, or if I was just training myself into an injury. Asking a stranger to watch me felt impossible. Hiring a personal trainer at $60 to $150 a session was not realistic on a student budget.

So I did what I always do when I cannot find something that solves my problem. I built it.

Once I started thinking about this I realized how many people face the exact same situation.

Physical therapy patients get discharged from hospital with a printed sheet of exercises and told to keep practicing at home. No feedback. No correction. No way to know if they are helping themselves or quietly making things worse.

People in smaller towns and rural areas often have no physical therapist or qualified trainer within reasonable distance. Working class people for whom $150 per session is genuinely not an option. Older adults trying to stay active independently who face the highest risk from poor movement but often have the least access to professional guidance.

All of these people have one thing in common. They are doing their best alone with no one watching.

FormForge AI is a training companion application powered by Amazon Nova's multimodal capabilities on AWS Bedrock.

It does four things.

The first is AI form analysis. You upload a short video or a few photos of yourself doing an exercise. The app extracts six frames evenly spaced across the full movement arc and sends all of them to Amazon Nova Lite alongside exercise-specific coaching cues. Nova looks at your movement as a complete sequence and gives you a numeric form score from 0 to 100, tells you what you are doing well, what needs fixing, and flags anything that looks like an injury risk.

The second is a workout logger that works completely without video. You just enter your exercise, sets, reps and weight and hit log. The app is useful as a daily training tracker on its own, which means you will actually use it every day instead of just when you happen to have a camera set up.

The third is a meal nutrition tracker. You photograph your breakfast, lunch and dinner. Nova identifies the food it can see, estimates your macros, and tells you how close you are to your daily protein goal. Every estimate comes with a confidence level because honesty matters more than false precision.

The fourth is a progress analytics section and a color coded monthly calendar showing every training day, green for good form, yellow for sessions needing work, red where injury risk was flagged.

Most people building exercise analysis tools try to analyze a single image. One frame, one pose, one frozen moment. That approach misses almost everything important about how a person actually moves.

A squat is not a pose. It is a sequence of positions changing over time. Whether your knees track correctly through the full range, whether your hips reach adequate depth, whether your back rounds at the bottom — none of these can be reliably seen in one frame.

FormForge solves this by sending six frames simultaneously to Amazon Nova in a single context window. The prompt instructs Nova to reason across all six frames as a movement sequence rather than describe each one independently. This lets Nova detect compensation patterns that only appear at specific points in the movement, things a single image would completely miss.

This is the core of what makes FormForge genuinely different. It is not just a wrapper around an image recognition API. It is using Nova's ability to reason across multiple images together — which is a real multimodal capability — to do something that matters.

Amazon Nova Lite made the economics work. Each full analysis costs a fraction of a cent. A person using the app every day for a month spends under one dollar in API costs. That is what makes a genuinely accessible health product possible.

I built this for the version of myself that stood in the corner of a gym this January not knowing if anything I was doing was right.

But I also built it for people with much higher stakes than my ego. Rehabilitation patients who cannot afford to get their exercises wrong. People in areas without access to professional guidance. Students and working people for whom coaching is a luxury they simply cannot afford.

The potential of this application with the right professional input is significant. Consulting with licensed physiotherapists to validate the exercise cue library. Working with certified strength coaches to develop proper scoring standards. Building referral pathways so the app knows when something is beyond AI guidance and connects the user with a real professional.

Amazon Nova 2 Sonic could eventually deliver voice feedback during workouts in real time — corrections you hear while you are moving rather than read afterward on a screen. That one addition would turn FormForge from a review tool into something closer to a live coaching companion.

The quality of the prompt matters as much as the model. Telling Nova to reason like a physical therapist across a complete movement sequence produces dramatically better output than simply asking what is wrong with the exercise. Domain knowledge in prompts is a real advantage.

Being honest about what the AI cannot do makes the product more useful, not less. Every nutrition estimate in FormForge comes with a confidence level and a note that actual values may vary. Users trust a tool that admits its limits more than one that pretends to be perfect.

And the most important lesson of all: building something that is genuinely useful without the AI is what separates a product from a demo.

I started this year wanting to take my health more seriously. I ended up building something I genuinely wish had existed when I walked into that gym in January.

If you are a physiotherapist, a strength coach, or someone who trains alone and wants better feedback then FormForge AI was built for you.

#AmazonNova #AWS #Bedrock #HealthTech #FormForgeAI