Mira
Mira is the visualization layer for thinking.
Core trick
Fans a spoken question out to four parallel Gemini sub-agents that each build part of an animated scene; follow-ups mutate that scene.
Description
Mira is the visualization layer for thinking. You speak any question, and four Gemini 3.5 Flash sub-agents fan out to build a studio-grade animated explanation in under 30 seconds, with follow-ups that morph the scene in real time. Writing compresses mental models into text, and rebuilding them mentally is work most people skip. Authors like Bartosz Ciechanowski and Grant Sanderson solve this with hand-crafted interactive explainers, but each one takes weeks to months. Mira automates the artisanal layer so any concept can get its own: how a phantom traffic jam emerges from nothing, how an AI generates an image from a sentence, why a bicycle actually stays upright. Speak a question, see the explanation.
README excerpt
# Mira > **The visualization layer for thinking.** Built at the Google I/O Hackathon, May 2026. Mira is a generative visualization engine. You speak an idea, and it produces a Ciechanowski/3Blue1Brown-style animated, narrated explanation of that concept on demand, in seconds. Parallel Gemini 3.5 Flash sub-agents generate a synchronized in-browser animation plus narration, and your follow-up questions morph the existing scene instead of starting over. Think of it as a creative tool for turning ideas into moving visuals, the same category as a generative-art engine, not an assistant that answers questions for you. I sketched this idea at the Google I/O Hackathon, May 2026, and then built it in the room, so this is no longer just notes. It runs end to end: you type or speak a query, a fan-out of Gemini 3.5 Flash agents plans the scene, writes the p5.js or three.js animation, and narrates it, all streamed live into the browser, with a Managed Agents API verifier checking the animation against the narration. Everything below is the original idea-stage thinking, kept as-is, because the open questions I wrote down that morning are mostly the ones I actually had to answer while building. --- ## The problem I'm chasing Most ideas live as text or equations, which is a compressed format that throws away the spatial, moving, cause-and-effect structure that made the idea legible in the first place. Visualization is how you get that structure back, but right now building it is artis