Suggest that Facebook users react to trending news topics. By breaking down articles into 200+ characteristics, the team ensured a high interaction rate from tens of millions of worldwide users, and generated millions in ad revenue.

Medical watchdogs say that only 70% of case notes meet quality/accuracy guidelines. The team developed an Agent that uses transcripts to generate case notes that match the medical institutions style. These AI generated case notes have a 92% audit acceptance rate, and give practitioners back dozens of hours per week to focus on patient care.

Designed an Agentic Framework that breaks complex problems spanning multiple data sources into bite-sized steps, and gave the Agents the ability to use tools such as SQL and Web Searches.

Designed a novel technique that lets low-skill translators create specialized models to answer narrow questions like "What was the score of the Lakers game?". These small language models cost .1% of larger models to operate, and users prefer their answers to larger models.