Besedo
Trust & Safety taught me to treat model quality, delivery systems, and operational feedback as one product problem across languages, media, customers, and changing abuse patterns.
What I worked on
- Built and operated multilingual NLP, computer vision, and similarity systems for global Trust & Safety products.
- Created shared MLOps foundations that shortened the path from customer evidence to production release.
- Progressed from Senior Data Scientist to Head of Data Science while remaining involved in architecture and difficult production work.
In production
- More than 130 production models processing tens of millions of content items per day.
- Led a team of up to eight engineers and reduced delivery cycles from months to weeks.
System: NLP · Computer vision · Similarity search · MLOps · Kubernetes
The technology has changed, but the engineering lesson carries into agent systems: operational feedback is part of quality, and scale exposes assumptions that prototypes hide.