Principal TPM in Meta's AI Infrastructure org, where I drove the first production migration of a recommendation workload to next-generation LLM inference.

Coordinated cross-org execution of Meta's first production migration of a recommendation workload to next-generation LLM inference — aligning serving, acceleration, hardware, and efficiency partners across NVIDIA H100/H200 and AMD MI350X. Drove two first-of-their-kind LLM launches on Meta's ranking and search surfaces: a generative recommendation pipeline and a Llama-based search-relevance model replacing a legacy BERT system.

Manage 5 TPMs in AI Data and Responsible Engineering, where access-friction reliability work cut interrupted data queries from 35% to 1.4%.

Stood up Meta's third-party dataset review process from scratch, partnering with Legal to define the workflow. Aligned 252 AI researchers and a 48-person cross-functional team, cleared the backlog to zero against hard launch deadlines, and earned acknowledgment in the Llama whitepaper. Supported LlamaFirewall — Meta's open-source guardrails for AI agents — and drove 59 novel GenAI risk reviews as the space scaled inside the company.

Restructured the fact-gathering function and vendor mix, cutting contractor cost roughly 90% while improving output quality.

Built Meta's post-Cambridge-Analytica API Governance program — the company's FTC consent-decree response — with oversight of API design, developer access controls, third-party app review, and regulator compliance for developer.facebook.com, one of the world's largest API platforms. Managed and led a team of 10 TPMs across a matrixed group of 159+ reviewers, and stayed technical enough to shape the direction of the platform projects rather than only running process.

The automation strategy behind it enabled 600+ product teams to ship third-party integrations roughly 90% faster, and was cited by Meta's Chief Product Officer as a company-wide benchmark for cross-functional execution. As Safeguard Owner for 20% of Meta's global privacy controls, consolidated 32 redundant safeguards and led FTC and EU regulator assessor responses.