A non-exhaustive view of enterprise systems I’ve led or built end-to-end — from on-site customer scoping through production rollout and monitoring — and the foundational research underneath them.
PCB Testing & Design
AI copilot for Electrical Engineers
Arena Physica · Principal AI Engineer
Most recent deployment, owned end-to-end across three pillars — core agentic architecture, the data layer, and the systematic-eval stack — built to make Electrical Engineers dramatically faster at PCB design and testing.
Approach: custom data structures for efficient agent context handling over PCB-native artifacts · systematic evals and self-improving agentic systems · close collaboration with domain EE experts · forward-deployment with users to showcase capabilities and iterate quickly
Data Centers
Anomaly detection at hyperscale
Arena Physica · Principal AI Engineer
End-to-end ownership of a high-priority anomaly-detection deployment over multi-node GPU and infrastructure telemetry — architecture, evaluation harness, and the on-site customer loop.
Approach: anomaly detection across unstructured server logs and time-series telemetry · agentic triage · multi-agent architecture · structured eval · efficient context handling via smart log compression and code execution · technical-documentation ingestion
Complex Hardware
Agentic systems for drones and flight systems
Arena Physica · Lead ML Scientist & Head of ML
LangGraph-based multi-agent architectures for autonomous issue identification and resolution across drone fleets and flight systems — built to make hardware and systems engineers dramatically faster on complex operational workflows.
Approach: multi-agent architecture · LangGraph · tool-use · code execution · retrieval · structured evaluation · production reliability patterns
Advanced Manufacturing
Yield optimization, anomaly alerts, tuning & validation
Arena Physica · Lead ML Scientist & Head of ML
A trio of deployments on advanced-manufacturing and hardware-validation lines: dynamic machine-parameter selection lifting yield by 3–5% on contact-lens production; physics-aware anomaly detection on liquid-aluminum casting that reduced false-alarm rates and surfaced defect modes operators had previously missed; and hyperparameter tuning for next-generation GPUs that accelerated multi-feature validation and compressed time-to-market.
Approach: physics-informed ML · deep learning · graph neural networks · contextual bandits · Bayesian optimization · time-series anomaly · model explainability · alerting pipelines
Post-Silicon Validation
Video artifact detection for high-performance GPUs
Arena Physica · Lead ML Scientist & Head of ML
Computer-vision pipelines that detect screen pixel-level artifacts and glitches in full-HD 60 Hz video output from high-performance gaming GPUs — deployed inside post-silicon feature optimization to accelerate validation and tuning of new silicon.
Approach: deep CV · distributed training · streaming inference · synthetic and real data generation pipelines
Revenue Systems · CPG
Dynamic pricing, recommendations, sales-force tasking
Arena Physica · Lead ML Scientist & Head of ML
Multi-arm bandit pricing, product recommendations, and sales-force action prioritization for CPG accounts — multi-$M annualized revenue uplift across deployments.
Approach: deep learning · recommendation systems · contextual & multi-arm bandits · uplift modeling · forecasting · simultaneous A/B testing of competing policies in production
AR Try-On
Real-time virtual try-on for eyewear
Ditto · Director of Research & Research Engineer · 2019 – 2021
3D face reconstruction, facial landmarking, face-shape classification, and PD/face-width estimation — deployed across web and mobile under strict latency budgets, lifting fit accuracy and conversion for partner retailers.
Approach: deep learning · 3D face reconstruction · landmark detection · classification · cross-platform inference
Patents: US11960146B2 (granted 2024) ·
US App. 20230360350
Semiconductor Fault Analysis
Super-resolution & CV in commercial inspection tools
Thermo Fisher · FEI · DCG Systems · 2015 – 2019
Productionized super-resolution, image fusion, CAD-to-image registration, denoising, and SEM object-detection algorithms inside commercial inspection frameworks used at the world’s leading fabs — methods that originated in my Ph.D. research at Boston University.
Approach: sparse coding · dictionary learning · image fusion · optical-system simulation
Patents: US20180293346 ·
US App. 20200333394Papers: ISTFA 2018, 2014, 2012 · Optics Express 2015 · DATE 2015 · ICASSP 2013
Foundational Research
Sparse-coding super-resolution & anomaly detection
Boston University · Ph.D. Research Assistant · 2009 – 2015
Cross-disciplinary research at the intersection of photonics, integrated circuits, and machine learning — sparse-coding and dictionary-learning methods for super-resolution in IC microscopy, biomass estimation in interferometric microscopy, and feature selection for anomaly detection in surveillance video. Some of the methods I developed here became the basis for productionized algorithms at DCG / FEI / Thermo Fisher and two US patents.
Labs: Information Sciences & Systems Lab · Optical Characterization & Nanophotonics Lab
Output: 6+ peer-reviewed papers (Optics Express, ISTFA, DATE, ICASSP) · foundation for two US patents