RectiFighter - A warrior against slouching

Help kids sit up straight while studying: RectiFighter tracks posture on-device, scores it like a game, and delivers personalized, Gemini-3–crafted AI feedback - without sharing photos or video.

Comment

Inspiration

The typical scenario is familiar: a first-grader starts doing homework - writing, gradually slouches, hunches their shoulders, and tilts their head to peek at the text from under their hand. It's uncomfortable, suboptimal for vision and breathing, and can become a long-term habit. Kids don't slouch because they "don't care" - they slouch because they're focused.

Parents want to help, but continuous monitoring is unrealistic. What if there were a friendly "coach" app that:

  • notices posture drift,
  • reacts consistently,
  • and motivates the child with game-like feedback?

Turning posture correction into a lightweight game plus an "AI voice companion" felt like a practical, kid-friendly solution.

And of course, I also wanted to try building an app entirely in Google AI Studio, putting Gemini 3 Pro Preview and Gemini 3 Flash Preview to the test in practice.

What it does

RectiFighter is a posture-coaching "mini-game" for kids doing homework. It monitors the student's sitting posture through a laptop or smartphone camera and turns good posture into reward points, while slouching costs points. When needed, it can also speak up with a short, humorous reminder - or some praise.

How I built it

Google AI Studio was my only tool: I described what I wanted in plain language, reviewed the result, iterated, or rolled back to a previous version - until I had this working proof of concept.

Initially, the idea was to periodically grab frames from the video stream and send them to Gemini 3 as images, asking it to estimate the degree of slouching. I had to abandon that approach due to unreliable evaluations, high cost, and the fact that many parents wouldn't agree to send photos of their children to the cloud for posture analysis. So I decided to run pose analysis locally on the device. Fortunately, MediaPipe Pose can detect useful landmarks from video.

In this project, Gemini 3 had a key role: finding the right words so that the child gladly corrects their subtly drifting posture and feels like they're part of a fun "good posture" game.

Challenges I ran into

At first, I had to adapt to the working style of Gemini 3 Flash Preview and Gemini 3 Pro Preview. These models are very proactive :-) and will happily "improve" parts of the program you didn't ask them to touch. Luckily, this is easy to address with a simple phrase like "don't make any other changes" at the end of the prompt.

The main technical challenges were identifying the most reliable landmarks for estimating slouching and finding the best MediaPipe Pose configuration.

Accomplishments that I'm proud of

I managed to build what I think is a fun and useful app in five evenings, while also learning several new tools and AI models.

What I learned

I learned Google AI Studio, tested Gemini 3 Flash Preview and Gemini 3 Pro Preview, learned to use MediaPipe Pose, and got familiar with PWAs.

What's next for RectiFighter - A warrior against slouching

The app isn't production-ready yet. Next, I plan to significantly expand the variety of rewards, polish the system prompts, and optionally add natural-sounding AI voice synthesis or use Gemini 2.5 TTS. I also need to decide how to host the PWA (a paid server-side API key vs. a user key stored on-device), along with many other smaller improvements.

Built With

Share this project:

Updates