Founding Engineer
CrazyScale
An AI video translation, dubbing, lip-sync, and browser-based editing platform built from the ground up.
- Launch
- Under 8 months
- Integrations
- 7+ providers
- Workflow
- End-to-end media
01 · Context
The system behind the product
Designed the core media engine and unified seven-plus AI and media providers behind a resilient product workflow.
02 · Architecture
How the system is shaped
- 01
A unified media workflow coordinates transcription, translation, voice cloning, dubbing, lip-sync, and final asset assembly.
- 02
Provider adapters isolate seven-plus AI and media services so critical stages can use fallback providers without rewriting the product workflow.
- 03
A browser editor exposes scene timing, waveform adjustments, audio correction, and proofreader feedback over the same underlying media model.
03 · Decisions
Trade-offs made explicit
Provider capability behind stable contracts
External AI services are treated as replaceable infrastructure, keeping provider outages and API changes away from the product model.
Human correction becomes product memory
Glossary-aware translation rules are learned from proofreader feedback and reused instead of being lost after each project.
Timing is part of translation quality
Duration matching, stem separation, and timeline editing are handled as first-class concerns rather than post-processing details.
04 · Outcome
What changed
Moved from concept to production launch in under eight months, including glossary-aware translation memory and duration-matched dubbing.
- Took the product from concept to production launch in under eight months.
- Unified seven-plus AI and media providers behind one resilient workflow.
- Preserved background audio and lip-sync quality through duration-aware dubbing and vocal/music separation.
05 · Capabilities
What the work involved
- AI video localization
- Provider redundancy
- Timeline editing
- Translation memory