Selected work
Senior data & AI engagements at United Airlines, Cisco Systems, Spotify, UHG Optum, Verizon, Wellstar, and the State of Connecticut. Plus three independent IP builds — DocSensei, ORCA, and Setup Score.
Engagements
2023 — Now
Lead enterprise AI, data, and analytics initiatives across 100+ state agencies, delivering measurable impact for 3.7 million citizens. Define the data strategy, ML roadmap, and governance posture that informs senior leadership and drives agency-level operating decisions. Standing up an enterprise LLM platform for document triage, policy lookup, and citizen-facing assistance.
State of Connecticut, DAS · Hartford, CT
2015 — 2023
Founded and led a specialized AI / ML consulting practice. Embedded senior data scientists and applied ML engineers inside Fortune 100 enterprises to deliver production ML, NLP, anomaly detection, fraud detection, content moderation, recommender systems, and enterprise analytics. Selected client engagements include United Airlines, Cisco Systems, Spotify, UHG Optum, Verizon, and Wellstar Health System.
Solution Cabin LLC · Atlanta, GA
2022 — 2023
Personalization and Marketplace teams: A/B and switchback tests for podcast and audiobook discovery, with power analysis under heavy-tailed engagement distributions. Built feature pipelines in PySpark on Databricks ingesting hundreds of millions of daily events. Prototyped an embedding-based content similarity service for cold-start podcast recommendations on FastAPI + FAISS.
Spotify · New York, NY
2021 — 2022
Network Assurance group. Unsupervised anomaly detection on streaming router telemetry using isolation forests and seasonal-hybrid ESD over Kafka topics. Designed a Bayesian change-point layer that reduced false-positive alert volume by an estimated 40% in pilot. Built a capacity-planning forecasting service combining Prophet and LightGBM, deployed via SageMaker behind an internal gRPC API.
Cisco Systems · San Jose, CA
2018 — 2021
Advanced analytics team supporting fleet operations and MRO. Built predictive maintenance models for line-replaceable units across narrow- and wide-body fleets, informing parts provisioning and turnaround scheduling at major hubs. Designed a digital-twin-style supply chain simulation in SimPy and PySpark, plus a Gurobi integer-programming model for crew and gate optimization under irregular operations.
United Airlines · Chicago, IL
Selected independent builds
DocSensei · 2023→
End-to-end RAG system for high-volume document workflows in insurance and legal. Hybrid retrieval (BM25 + dense embeddings), cross-encoder re-ranking, page-faithful citations, and a hallucination guardrail scoring each generated span against retrieved evidence. Architecture is directly applicable to commercial real estate loan files: rent rolls, operating statements, leases, intercreditor agreements, and trustee reports. Currently in beta at a U.S. insurance brokerage under a perpetual source-code license.
ORCA · 2024
Asset-level liability and exposure model for a national operator. Combined satellite-derived signals, regulatory schedules, and operator-submitted data into a single risk index. Quantified expected exposure under alternative regulatory regimes and surfaced under-reporting patterns. Geospatial layer integrates raster signals with operator records using PostGIS and Tableau map layers.
Setup Score · 2022→
Multi-factor signal generator (Setup Score v6 and v7) blending momentum, volatility regime, options flow, and earnings drift. Pine Script and Python research stack with live tracking across selected positions. Hands-on financial markets and instrument-level research, informing personal trading capital.
Private repositories. Demos available on request.
Capabilities
Modelling
Risk and propensity models, time series forecasting, survival analysis, NLP classification, structured prediction, and experimentation. Python first, with R where it earns its place.
LLM systems
Production RAG, structured extraction, evaluation harnesses, guardrails, multi-agent orchestration, cost and latency observability. Vendor-agnostic, biased toward what survives.
Data platform
Ingestion, transformation, semantic layer, governance, and the dashboard surface. Built around dbt, Airflow, Snowflake, BigQuery, Postgres and a clear contract layer between them.
Engineering
FastAPI, SQLAlchemy, Celery, Redis, Postgres, Docker. Patterns I have shipped to production at scale, with the operational telemetry that lets a team sleep through the weekend.
Healthcare
HIPAA controls, BAA scoping, PHI handling, audit logs, and the documentation a compliance officer will actually sign off on. Predictive readmission and care management models.
Leadership
Hiring rubrics, interview loops, IC ladders, performance frameworks, and the operating cadence that gets a team to predictable shipping inside ninety days.
A working note
I take on a small number of engagements a year. Fit matters more than fee.