Learnimi

Turn a favorite toy into a lively learning buddy.

Comment

Learnimi turns a photo of your favorite toy into your best learning companion that demonstrates safety and emotional skills through short animated adventures!

Judges can instantly try the public demo with pre-made Cappy or upload a photo for built-in lesson movements, with the photo remaining in the browser. Running Learnimi locally with a server-side OpenAI API key unlocks the complete supervised AI pipeline, including safety checks, background removal, and custom movement videos.

Inspiration

Safety lessons can feel abstract or boring, especially for kids. I wanted to make each action easier to understand and remember by letting a familiar and cozy companion show what to do.

What it does

A grown-up can approve a photo of a toy, create a learning buddy for the kid, and choose from four adventures:

  • Road safety
  • Online safety
  • Home safety
  • Big feelings

The same companion stays with the learner throughout the journey, demonstrates each action, reacts to quiz answers, and celebrates completed lessons.

How I built it

Learnimi combines a responsive React and TypeScript interface with an Express media pipeline. Images are guardrailed, normalized and stripped of metadata before evaluation. OpenAI moderation and GPT-5.6 provide two fail-closed safety boundaries around the optional generated-buddy flow. GPT Image creates the isolated companion, course-specific animation is prepared in parallel, and completed media is stored locally in IndexedDB.

NOTE! The original upload is NOT persisted by Learnimi.

How Codex and GPT-5.6 helped

Codex was my collaborative engineering environment from the first playable prototype through the finished submission. It also helped turn browser feedback (Playwright) into focused React and CSS improvements, structure the guarded media, stabilize playback and recovery behavior, and build automated verification for privacy, responsive layouts, and media continuity.

GPT-5.6 is the structured vision reasoning layer at two safety boundaries. It evaluates the normalized source alongside image moderation, then checks the isolated companion before animation can begin. Unavailable, malformed, ambiguous, or rejected results stop the uploaded-buddy flow.

Challenges

The hardest part was keeping one companion visually continuous while making its movement meaningful for different lessons. I also had to coordinate multiple media jobs, recover cleanly from interruptions, preserve a fast no-key judging path, and keep child safety and privacy fail closed. Safety and security is crucial!

What I am proud of

  • Learnimi will help kids to learn in a fun, responsible and most importantly, safe way
  • One vivid and cute animated companion visibly demonstrates a bounded learning curriculum
  • A complete, responsive journey that works best for kids
  • Motion and sound controls with accessible interaction
  • Safe cancellation, recovery, cleanup, and local persistence
  • Comprehensive automated tests plus lint and production-build verification

What is next

The next step is testiing the learning experience with parents and educators, strengthening the production media lifecycle, and expanding the curriculum while keeping every upload behind explicit grown-up approval!

Built With

Share this project:

Updates