Flagship work
Eight production tools inside a large enterprise
The largest body of work on this site. Eight AI and automation tools shipped in 2026 into a clinical utilization-management workflow — built, packaged, documented and supported through 20+ releases on the flagship alone.
Each one followed the same loop: observe the workflow, define the requirement and the measure, prototype in a week, validate against live cases with the people who would use it, harden it, document it, train the users, then iterate from their feedback.
Reach differs by tool and is stated plainly rather than rounded up. OnePA Doc Alert is the only one deployed to the whole team — roughly 60 pharmacists use it daily. PAHub Helper and the bulk automation are used weekly to monthly, by me and a small group. The rest are single-operator or small-group tools.
Built from inside the workflow
Every requirement came from a case worked first-hand rather than from a requirements document, which is why the verdict states and the edge cases match what actually comes up.
The reviewer stays the decision-maker
No tool makes a determination. Each removes a lookup, a navigation step or a transcription, and surfaces the signal that drove its guidance so the reasoning can be checked rather than trusted.
De-identified by design
No client data leaves the environment. PHI handling shaped the architecture from the first prototype rather than being retrofitted.
Packaged, documented, supported
Distributed from a team library with first-run self-setup and no auto-sync dependency. Installation guide, user guide, banner reference and SOP written alongside the code; regressions fixed from user reports within days; changelog kept in the file.
Four of the eight, documented
Also built across the suite
- Auto-advancing clipboard automation with global keyboard detection for rapid question-set assembly
- Real-time conditional-logic navigation that highlights the live decision path through branching criteria
- Excel-to-interactive-HTML conversion for complex approval and denial logic
- Criteria indexing across ~100,000 workbook rows, resolved in seconds at the point of use
Evaluation work on an enterprise AI platform
Separately from the tools above: since March 2025 I have been part of a small group evaluating an enterprise AI review platform, starting with a week-long working session on its first draft before it was a product.
The work is evaluation engineering — grading AI-generated guidance case by case against source criteria, logging defects and failure modes back to the product team, and building the training material for launch. I am in the pilot group for the current agentic release, which moves beyond LLM processing of documents to agent-driven steps.
It is the same discipline as the verification harness on the ML platform: decide what correct means, measure against it, and report the failures rather than the wins.
Enterprise Prompt Engineering certificationEnterprise GenAI Generalist certification
These tools are actively deployed inside a Fortune 20 enterprise environment, so they cannot be fully demonstrated in public and the employer is not named here. De-identified throughout; no client data leaves the environment.