From research to production.

AI/ML Engineer with 5+ years shipping production systems across healthcare, automotive, and fintech. Currently a Graduate Research Assistant at University of Maryland advised by Dr. Alan Sussman (Sep 2026-present), on HPC I/O research for Open OnDemand in collaboration with the Ohio Supercomputer Center and Princeton Plasma Physics Lab (PPPL). Previously GRA with UMD's Research & Development department (Aug 2025-Sep 2026).

Recent roles: AI/ML Software Engineer II at Aya Healthcare: built AI-driven provider matching and demand forecasting for healthcare staffing (reduced placement time via intelligent skill-to-role matching), and shipped multi-agent LLM platforms scaling to 2.3M+ patient profiles at 95% clinical accuracy; Software Engineer (ML) Intern at AWS Redshift (vector search infrastructure prototype, production AI/LLM pipeline integrations, Summer 2026).

AWS Redshift ML Infrastructure · Summer 2026

• Amazon AUTA AI Hackathon Org-wide top winner among 800+ Amazon interns for a multi-model cost-aware router that cut inference cost 3.2x with only 1.8% accuracy tradeoff
• Native Vector Search Shipped a prototype that brings semantic search directly into Redshift's query engine. Sub-100ms latency on 500K-50M vector benchmarks, matching specialized vector DBs while cutting operational overhead of external services
• Bedrock ML Integrations Solved two pain points for Bedrock's integration with Redshift ML: in-warehouse embedding generation and state management for long-running inference pipelines. Cut LLM time-to-production from weeks to days

Prototype brought to near-production readiness. At Redshift's scale, building reliable infrastructure around the model is the hard part. Shipped a working system adopted by cross-org teams.

AI/ML Software Engineer II - healthcare staffing platforms and clinical LLM systems at national scale.

• Multi-Agent LLM Platform Clinical decision-support system on 2.3M+ patient profiles at 95% clinical accuracy
• Provider Matching Engine AI-driven skill-to-role matching that reduced healthcare-staffing placement time in production
• Demand Forecasting Real-time forecasting pipeline for staffing demand across US healthcare networks

Google Summer of Code 2026 - The Linux Foundation

Completed GSoC 2026 (May-Aug) on the Accord Project's APAP / MCP server, mentored by Niall Roche and Dan Selman. Built a protocol layer connecting AI agents (Claude, ChatGPT) to smart legal contracts. Delivered a service-layer refactor with four-tier testing (90%+ coverage) and POC with 53 tests, 98.55% statement coverage.

Research output: Framework Wars Are Cheaper Than Most People Think - benchmarked MCP vs OpenAI function calling vs LangGraph. Key finding: tool definition overhead dominates (1,100+ tokens), making selective tool exposure the real cost lever. Full case study →

Judge and mentor at HackMIT 2025 and LA Hacks 2026 (1000+ builders, 36-hour hackathons at MIT and UCLA). Coached emerging talent through rapid prototyping under pressure, evaluated projects across AI, infrastructure, and systems domains, and worked with co-judges and organizers to identify technical depth and real-world problem-solving approach over flashiness.

At Bridgestone Group (Azuga Inc.), led cloud operations, data architecture and MLOps on fleet telematics at planetary scale.

• Data Infrastructure 24 billion sensor records across 120 PB warehouse, processed with sub-60s latency on Apache Spark 24-node clusters
• Accident Risk Survival Model Cox regression on 7.8M crash records (C-index 0.78) projecting $122.9M+ savings
• Streaming Warehouse Real-time pipeline at 99.8% data accuracy, NLP sentiment systems, and a Generative AI "Chat with Data" initiative on LangChain + AutoGen

30+ production projects on GitHub. Featured systems include:

• AgentForge Multi-Agent RAG Platform - 98.5% success rate on 10K+ concurrent requests, LangGraph orchestration with adaptive routing
• Real-Time Recommendation Engine - sub-100ms latency, 500K+ daily predictions on PySpark + Delta Lake
• Driver Behavior Forecasting - Bayesian hierarchical modeling, shipped at Bridgestone (7.8M crash records, C-index 0.78)
• Additional portfolio - Demand forecasting (MAPE 2.8%), Thompson sampling experimentation platforms, optimization engines (97.2% matching efficiency)

Published in IEEE and IJAET on predictive maintenance and supply-chain optimization. Technical articles on Medium / AWS Tips Publication and technical deep dives on LLM frameworks and RL systems. See all publications →

I focus on building reliable, production-ready AI - from HIPAA-compliant healthcare systems to real-time fraud detection processing 500K+ daily transactions at 94.2% accuracy. The aim is always the same: scalable systems that deliver measurable business impact under real-world constraints.