OP here. A few additional technical details folks might be curious about:
- Why Gemini over GPT-5/Claude4.5 for agent orchestration: Gemini3.0 is better for react code.
- Visual Box Engine specifics: ~300 parameterized animation templates. Each "box" is a React component with props like {concept, relationships, emphasis, timing}. Example: "CauseEffectFlow" takes an array of steps and auto-generates animated arrows + state transitions.
- E2B sandboxing: We run Remotion preview renders in isolated environments. This prevents malicious/buggy code from affecting other users' jobs.
Happy to answer questions about any part of the stack!
OP here. A few additional technical details folks might be curious about:
- Why Gemini over GPT-5/Claude4.5 for agent orchestration: Gemini3.0 is better for react code.
- Visual Box Engine specifics: ~300 parameterized animation templates. Each "box" is a React component with props like {concept, relationships, emphasis, timing}. Example: "CauseEffectFlow" takes an array of steps and auto-generates animated arrows + state transitions.
- E2B sandboxing: We run Remotion preview renders in isolated environments. This prevents malicious/buggy code from affecting other users' jobs.
Happy to answer questions about any part of the stack!