ML Tech Lead & Applied AI Researcher
Building production AI at Instacart — where my work on personalized fulfillment intelligence has driven $200M+ in influenced GTV and is cited in company shareholder letters. Previously at Samsung R&D and Walmart. Research cited 127+ times. 4 patents.
I'm a Machine Learning Tech Lead and applied AI researcher known for designing and shipping large-scale recommendation, retrieval, and personalization systems. At Instacart, I've led the ML architecture for personalized product substitution — a capability publicly credited in shareholder letters for improving customer satisfaction and repeat orders. These systems handle hundreds of millions of item replacements annually at sub-100ms latency.
My career spans Samsung R&D, where I built on-device AI for flagship Galaxy devices, and Walmart, where I built grocery recommendation systems at scale. I hold an MS in Computer Science from UMass Amherst, where I published research under Prof. Andrew McCallum in collaboration with Goldman Sachs.
Member of the Forbes Technology Council, IEEE Senior Member, and International Academy of Digital Arts & Sciences. Invited reviewer for AAAI, KDD, ACL, WWW, and UMAP — 12+ program committees.
Sole ML engineer and de facto technical lead for Instacart's business-critical substitution ranking system. Evolved through four major architectural phases — multi-stage retrieval + deep ranking, explicit negative event modeling, personalized multi-surface ranking, and real-time re-architecture achieving ~11ms p99 latency. LLM-curated denylists and semantic retrieval eliminated worst-case failures. Work cited in Q2 & Q4 2025 and Q4 2024 Instacart Shareholder Letters. Individual Tech Achievement Award, Q4 2024.
Founding member of Samsung's on-device AI team. Designed a FlatBuffers-based serialization architecture for dictionary tries that cut first-query latency 40× (2s → 50ms). Implemented Rocchio query expansion and TF-IDF semantic matching for a 25% recall improvement. Built on-device LTR models (RankSVM, LambdaMART) reused for Bixby card and notification ranking. Shipped on Galaxy S10+ and subsequent flagships — 5M queries/day. Published 4 peer-reviewed papers.
Under Prof. Andrew McCallum, built a GPT-2-based saliency classifier for low-resource long document summarization, outperforming strong baselines by +6 ROUGE points (ACL 2021, 83 citations — cited by Google Research at NeurIPS 2023). Proposed a nearest-neighbors relation prediction method for scientific text, outperforming prior work by up to +3 F1 points (EMNLP 2020). Applied in collaboration with Goldman Sachs for financial document NLP.
Open to collaborations,
advisory, and speaking.
The best way to reach me is by email.