01
The challenge
The production pipeline was overusing expensive animations, producing inconsistent visuals, and had lost its Manim rendering capability.
Multi-agent systems · Educational AI
Interactive Course Agents
A multi-agent platform that turns narrated lessons into interactive courses with matched visuals, layouts, animations, and synchronized overlays.
1,300+ tests kept green
Implementation details presented without public source code.
Video case study02
Engineering approach
I introduced deterministic representation routing, type-aware planning, a tested animation component library, and a self-correcting author-to-render repair loop with budgets and graceful degradation.
03
Selected outcomes
- Restored a balanced mix of visual representations.
- Drove broken animation renders to approximately zero.
- Recovered the production Manim pipeline.
- Eliminated deployment race conditions while preserving more than 1,300 passing tests.
Next project