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.
Background: each dot stands for an equal share of real OSHA ITA filings, 2016 to 2024. The 2.07% that swell carry 96.7% of all reported hours, drawn to scale by area. The finding
How running an EHS system turned into a research programme. One question carries through: can you trust the number?
As sole global administrator of the enterprise EHS system, I own the path from an incident entered at a site to the TRIR on a board slide, and the OSHA filing behind it. The weak point is rarely the injury count. It is the hours worked underneath it.
Medallion pipelines in Databricks under Unity Catalog, then a Foundry ontology spanning EHS, supply chain, procurement and finance. Every figure traces back to a row. See the architecture.
Natural-language agents over live systems through MCP, with deterministic calculations doing the arithmetic so the same question returns the same number. Read the case study.
Across 2.8M public OSHA filings, 2.07% of filings carry 96.7% of reported hours, and screening implausible hours moves the all-filer ITA TRIR by 1.4x to 249x depending on the year. The same test now applies to AI answers: grounded in an authoritative source, or not. See the research.
Before a TRIR or an AI answer reaches a board slide, it should carry its hours denominator and its source. That is the standard I build the platforms to, and the one the research tests in public. Read the case studies.
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.
Gensuite (Benchmark ESG)
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.
Python, linear programming
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.
Databricks, PySpark, Unity Catalog
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.
TRIR per 200,000 hours worked, by calendar year, CY2016 to 2024
TRIR = recordable cases × 200,000 / hours worked, aggregated by year across OSHA Form 300A summaries submitted through ITA, CY2016 to 2024. Screened drops filings reporting under 120 or over 4,500 hours per employee. ITA covers establishments with 250+ employees, and 20 to 249 in designated industries, so small sites and exempt industries are absent. Source: ehs-osha-analysis outputs/summary.json. Hover or tab across points for values.
| Year | Screened TRIR | Unscreened TRIR |
|---|---|---|
| 2016 | 3.921 | 2.729 |
| 2017 | 3.998 | 2.634 |
| 2018 | 4.069 | 2.920 |
| 2019 | 4.018 | 0.016 |
| 2020 | 4.491 | 2.792 |
| 2021 | 4.368 | 2.385 |
| 2022 | 4.605 | 2.661 |
| 2023 | 3.147 | 1.380 |
| 2024 | 3.642 | 0.280 |
Strip those filings and the TRIR correction swings from 1.4x to 249x between years. It is not a portable constant, so one correction factor applied across years would misstate the rate. This is a data-quality finding, not a claim about why injuries changed.
Not the same panel as RPT-005’s 1.53% / 87% (CY2023-2025): this figure covers CY2016-2024, and the two use different filings and screens. Sources: ehs-osha-analysis · ehs-benchmarks.
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: 823 tests passing, screened TRIR 3.98.
Seven systems, from an enterprise ontology to a live public app.
All case studiesFrom CRM reporting to owning a chemicals manufacturer's global safety system in six years.
Every diagram here responds to your input. RPT-001 to 004 run on synthetic demo data; RPT-005 runs on 1.18 million actual OSHA filings. RPT-006 maps each public repository to a cited federal interest.
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 five times, three of them with a STAR Award.
Selected by the CIO, VP and General Manager for a two-week international IT leadership assignment, assessing every data flow at a 24/7 plant in the Port of Rotterdam.
Built a Python linear-programming engine that ranks Hexion’s micro markets by competitive opportunity, replacing market-by-market judgment calls with one repeatable model.
Designed the Account 360 ontology and shipped an AIP chat assistant and an OSDK app on it: one view of every account across CRM, SAP finance, supply chain, proactive alerts and Icertis contract lineage.
Designed a document ingestion and maintenance algorithm built on a traditional API workaround, keeping the document store current at a fraction of the API cost.
Designed and implemented, end to end, the metrics behind the site safety incentive program that sites are measured and rewarded on.
The statistics kept up with the job; a PhD in applied AI is next.
Newest first. Verify links open the issuer's record.