Cost avoidance from RLAIF/MoE eval automation
$100M
Eval lead time
7d → 24h
Dogfooding participation
30% → 90% in 3 months
- Own end-to-end Contextual and Proactive AI program strategy across Meta's smart glasses portfolio—orchestrating on-frame inference, cloud LLM serving, multimodal perception, and context understanding to deliver agentic, ambient AI experiences powered by frontier models that define the next generation of consumer AI devices.
- Drive the end-to-end RLHF data and evaluation strategy for Wearables AI, building human-in-the-loop infrastructure for model quality measurement, error classification, and training loop optimization—driving weekly VP-level quality reviews. Launched an RLAIF initiative on Mixture-of-Experts architecture that automated evaluation across 100% of wearables traffic, compressing lead time from 7 days to 24 hours and delivering $100M in cost avoidance.
- Pioneered the Wearables dogfooding measurement program in partnership with SVP leadership—building the end-to-end data pipeline, instrumenting on-device telemetry, and delivering an executive dashboard that established organizational visibility into real-world device usage for the first time. Drove participation from a 30% baseline to over 90% within three months.
- Champion cross-Meta AI adoption initiatives within the TPM community—building agent-based tooling, delivering technical talks, and partnering with leadership to establish AI as a force multiplier for program execution.
On-device inferenceCloud LLM servingMultimodal perceptionRLHFRLAIFMixture-of-ExpertsAgentic AI