Applied models, technical frameworks, and field-deployable tools.
A client-side validation engine that tests EHS performance metrics for predictive validity. By running lag correlation analysis on monthly metric uploads, it algorithmically separates true leading indicators (those providing operational signal before an event) from compliance metrics that merely count what already went wrong.
A suite of targeted extraction frameworks designed to eliminate administrative friction in safety workflows. By using schema-constrained LLM parsing, these prototypes automate the conversion of chaotic EHS text—incident narratives, SDS data, and regulatory clauses—into audit-ready JSON payloads. It provides the technical blueprint for moving from manual data entry to automated data refineries.
An interactive behavioral economics engine modeling Stop Work Authority decision-making. Rejecting the assumption of the 'rational worker,' the model applies Prospect Theory and Hyperbolic Discounting to mathematically calculate why field crews avoid utilizing SWA. It maps safety culture as a quantifiable algorithm rather than a lagging observation.
Run Simulator →A Natural Language Processing (NLP) pipeline decoding the professional identity of the safety industry. Ingesting unstructured practitioner responses, the model uses SBERT sentence embeddings, K-Means clustering, and UMAP to map semantic archetypes. It algorithmically quantifies the gap between compliance metrics (ISO 45001) and operational resilience (Safety-II) using verb classification models.
View Analysis →A production Chrome extension that modifies the osha.gov DOM to eliminate information friction. It automatically detects and embeds official OSHA Letters of Interpretation directly adjacent to the target standard, and maps regulatory cross-references (e.g., 29 CFR 1910.134) in real-time. It tightens the feedback loop between monolithic compliance text and operational queries.
Chrome Store →A machine learning pipeline evaluating Logistic Regression, XGBoost, and BERT models to automatically classify unstructured workplace injury narratives. Trained on a decade of OSHA Severe Injury Reports (2015–2025), the application parses extreme events (amputations, hospitalizations) against the OIICS v2.01 standard (Nature, Part, Source, Event). It demonstrates how to pull quantitative signal from free-text field data.
View Application →A native iOS engine for field engineers calculating dropped object risk (DROPS). The model computes potential impact energy (Joules) via mass and drop height, mapping the physical output to structural risk classifications. It bypasses spreadsheet-based field notes by generating standardized assessment data and compliance-ready PDF reports directly on-site.
App Store →