Jun 2020 – Present · 6 yrs · New York, United States

• Technical lead for foundational AI infrastructure powering Meta AI Search Engine (MASE), supporting ~50M QPS and ~50B documents inventory
• Architected large‐scale retrieval and ranking platforms using KNN Search, Product Quantization (PQ), and embedding‐based retrieval to enable high‐throughput, low‐latency ML inference in production
• Led development of Visual Search for Instagram, processing ~10M media assets per day, including:
• ML classifiers to determine media appropriateness and safety
• Vision models to identify shoppable products in images, videos
• Video pipelines to extract smart keyframes for ML tasks
• Designed and deployed multimodal retrieval systems leveraging visual and contextual embeddings to retrieve visually similar products from a ~10B inventory
• Implemented write‐through caching and capacity‐aware serving strategies, significantly reducing infrastructure utilization while maintaining strict latency SLOs
• Built and operationalized AI agents to automate issue triage and resolution surfaced during large-scale dogfooding, reducing manual intervention and accelerating iteration cycles
• Fine‐tuned ML models and LLM‐based agents using prompt engineering to measure relevance, prominence, and ranking quality, improving business metrics like CTR, STR, user engagement
• Drove end‐to‐end performance, reliability, and cost optimization initiatives, achieving 10× scalability improvements and sustained 99.99% service availability
• Created capacity models and Virtual Resource Templates (VRTs) to forecast compute, memory, and storage needs across heterogeneous hardware profiles with cost optimizations
• Served as a cross‐org technical leader, aligning Facebook, Instagram, Product, Platform, and Infrastructure teams to define short‐term execution plans and multi‐year technical roadmaps
• Mentored and guided 25+ senior engineers across Infra, Product, and ML teams, raising the technical bar through architecture reviews, design standards