← All work

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

  1. 01

    A unified media workflow coordinates transcription, translation, voice cloning, dubbing, lip-sync, and final asset assembly.

  2. 02

    Provider adapters isolate seven-plus AI and media services so critical stages can use fallback providers without rewriting the product workflow.

  3. 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