San Francisco Bay Area
I build AI systems to stop adversarial abuse at Google scale. My focus is on Scalable Oversight, building: - Agentic Evals: Agents that identify novel attack patterns and infer adversarial intent - Adversarial Robustness: Systems that detect and react quickly to fast-evolving threats I approach safety with a system reliability mindset, as a former hardware engineer (RF).
- Agentic Scalable Oversight: Architecting LLM-based agent systems for measurement and detection, replacing manual reviews with high-throughput AI evaluators - Adversarial Robustness: Increasing attacker costs by requiring multi-signal coherence for each attack - Measurement / Alignment: Created metrics and frameworks to ensure visibility on defensive gaps and impact to good advertisers
- Ecosystem Integrity: Mitigated engagement abuse volume by 30x - Anti-abuse SME: Led safety requirements for high-stakes launches (Jewels, Hype), acting as the core Subject Matter Expert for financial abuse and engagement integrity - ML Detection: Deployed embeddings and weak supervision models to neutralize invalid engagements at scale
Pinecone is a fully-managed vector database for similarity and relevance ranking (product recommendations, ads, jobs, dating). Pinecone enables ML and infrastructure engineers quickly build, launch, and scale vector-based ML applications, utilizing recent applied science findings.
Branded consumer experiences for eCommerce brands - 3x improvement in accuracy of package delivery date forecast - Led overhaul of customer-facing and internal data analytics interfaces
HIPAA-compliant mobile communications for caregivers to seniors - Grew paying customers by 5x - Led releases for multiple app versions on iOS, Android, and web - Built web console (ReactJS)