I turn plant, safety, and operational data into decisions executives can act on.
I own the enterprise EHS management system for a global chemicals manufacturer end to end, build the data and AI platforms underneath it, and lead the people who ship it.
Safety numbers are only as good as the denominator and the citation underneath them.
Three systems now running in daily use at Hexion.
Every global site needed a modernized way to capture safety data inside an aging enterprise EHS application.
Led an end-to-end refresh of the application across two phases, from requirements through global rollout.
80+ feature enhancements shipped, reshaping how every global site captures safety data.
Leadership needed to see where competitive opportunity sat across dozens of micro markets, without a shared model.
A Python linear-programming engine that scores and ranks those markets.
Adopted company-wide and awarded the Transformation Award.
Contract, ERP, and warehouse data sat in separate systems nobody could audit together.
Databricks medallion pipelines unifying Icertis, Dynamics 365, SAP, and Gensuite feeds.
Migrated the platform to Unity Catalog governance, making every downstream report audit-ready.
Real OSHA data says the screening step every EHS team assumes is optional actually decides the number.
Unscreened TRIR collapses to 0.016 in 2019 and 0.280 in 2024. The screened series stays between 3.1 and 4.6 across all nine years.
Why it matters: a benchmark or peer-group comparison built on the unscreened series is reading a data artifact, not a safety trend.
Source: ehs-osha-analysis, committed summary of screened and unscreened TRIR by year. Hover or tab across points for values.
Strip those filings and the TRIR correction swings from 1.4x to 249x between years. It is not a portable constant, and any benchmark that treats it as one is wrong by an amount nobody has measured.
A draft paper, six public repositories with their own sites, and a plain statement of what has and has not been established. Real OSHA data where claimed, simulation where labelled.
Programme status: 810 tests passing, screened TRIR 3.98.
Six systems shipped, from an enterprise ontology to a live public app.
Leading the company's migration to Palantir Foundry - a single ontology and digital twin spanning EHS, supply chain, procurement, and finance, replacing scattered reporting with one governed model.
A Python linear-programming engine that pinpoints competitive opportunities across micro markets. Earned the company's Transformation Award.
Led an end-to-end refresh of the enterprise EHS application, delivering 80+ feature enhancements that reshaped how every global site captures safety data.
Natural-language-query and summarization agents on Copilot Studio, Claude Code, and Codex, with an MCP architecture grounding them in live systems and deterministic algorithms keeping answers reproducible.
Production Databricks pipelines unifying Icertis contracts, a 40-entity Dynamics 365 model, SAP, and Gensuite warehouse feeds into audit-ready reporting - then migrated wholesale to Unity Catalog governance.
A live decision-support app tracking public operating signals for chemicals, resins, engineered wood, and construction demand - EPA, Census, and BLS data turned into scored, tiered state-level actions and scenario models.
From CRM reporting to owning a chemicals manufacturer's global safety system in six years.
Every diagram here responds to real input, including 1.18 million actual OSHA filings.
Four disciplines, each owned end to end.
Global owner of the enterprise EHS management system and the regulatory reporting that depends on it.
Production pipelines that turn scattered source systems into governed, audit-ready reporting.
Agents that let people ask questions of enterprise data in plain language - grounded, auditable, cost-aware.
Board-level reporting, safety strategy, and optimization models used by leadership across the business.
Leadership singled this work out three separate times.
International IT leadership assignment - Pernis site, Port of Rotterdam, Netherlands
Micro Market Model - Python LP optimization for competitive market opportunity
End-to-end design and implementation of site safety incentive program metrics
The statistics kept up with the job; a PhD in applied AI is next.