# I-am-Solo

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [Rasika Deodhar](https://cerebralvalley.ai/u/rd95)
- **GitHub:** https://github.com/Rasika-Deodhar/sketchbridge
- **Demo video:** https://drive.google.com/file/d/10cQzp1l6jx1U6yOEJzH_DPM6VP-6lxqg/view?usp=drive_link
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1174

Sketchbridge turns a hard concept into a picture analogy, checks its own drawing, and then checks that the learner actually understood.

A learner asks a question by typing or by voice ("What is backpropagation?") and picks a world they know ("Restaurant"). Muse Spark proposes 3 analogies. The learner picks one, and Muse Spark writes it with an explicit mapping (e.g. "tasting the too-salty soup" = the loss function). Muse Image draws the scene, and Muse Spark looks at the drawing and redraws once if it doesn't show the mapping.

Then the learning loop:
• Connect the dots: Muse Spark locates every part of the analogy in the picture and walks through them in the order the concept actually works, with numbered dots, an animated path, and a one-paragraph "whole picture".
• Show the next scene: close-ups of individual parts in the same art style, each linked to the scene before it.
• Explain it back: the learner explains the idea in their own words (typed or spoken). Muse Spark grades each part as got / partly / missed, flags misconceptions such as taking the analogy literally, and offers a scene aimed at the biggest gap.
• I don't get it: a simpler explanation and simpler picture that keep the core idea.

The problem: explanations are one-way, and learners often don't know what they missed. Sketchbridge closes that loop with retrieval practice, and makes visual explanations for any concept for a few cents each (about $0.03–0.05 per illustrated concept).

It's a FastAPI app with a framework-free web page, deployed on Render with GitHub Actions CI. 5 ready-made examples work with no key; other requests sit behind an access code to protect API credit

## More from Meta Global AI Developer Hackathon

- [ManualMate](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1171)
- [Latent Futures](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1172)
- [Fahaam Qureshi](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1173)
- [WendstormXR](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1175)
- [Groundwork](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1176)
- [Team Invincibles](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1177)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1174. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt