I work the seam where clinical operations become revenue, and I build analytics that prove their own numbers: every metric on this page regenerates from a repo you can clone and run.
| ⌛ 3+ years Epic HB / Resolute across large health systems |
🧪 150+ scenarios UAT executed for an Epic upgrade cycle |
📊 5K+ daily txns supported, manual reporting effort cut 30% |
🏆 4 certifications Johns Hopkins ×2, MedCerts ×2 |
📦 4 repos every benchmark measured, 95–96% test coverage |
Four repos, four different revenue cycle bottlenecks, four different mechanisms. Each includes a labeled synthetic data generator, so detection quality is a measurement against ground truth, not a claim.
| Repo | What it proves | Headline number |
|---|---|---|
| claims-denial-leakage-miner | Preventable denials can be classified to an actionable root cause even when remit codes lie | 99.5% precision, 100% recall, 99.1% cause accuracy; $5.57M of $7.28M denied dollars traced to six preventable causes; 126K claims/s |
| rcm-upgrade-regression-sentinel | Upgrade UAT can be a declared contract with a measured catch rate, not an eyeball exercise | 100% catch rate over 1,299 planted regressions, zero false alarms on a benign-only control; 1M rows in 67s; generates the Excel sign-off workbook |
| hl7-charge-capture-reconciler | Ordered care that never becomes a posted charge is findable and priceable from raw HL7v2 | 0.98 missing-charge recall with honest false positive accounting; ~30K msg/s; $1.8M missing and $2.8M late charges priced |
| workqueue-flow-radar | Conflicting routing rules that ping-pong claims between queues can be caught and named from the event log alone | 1.0 precision and recall on labeled victims; 1.1M events in 7s; renders the daily ops packet (Excel + PDF) |
- Turning denial reports into root-cause worklists a biller can act on without asking why a claim was flagged
- Automating UAT evidence for EHR upgrade cycles: the tolerance spec is the test plan, the workbook is the artifact
- Workqueue flow analytics: aging, first-pass yield, and the routing conflicts native queue views cannot see
- Porting the repos' plain-SQL models to MS SQL Server syntax, the warehouse most hospital reporting teams run
I learned the revenue cycle from inside its workflows: UAT scripts, workqueues, and production support tickets at hospital scale. Then I went back to school for the data discipline (MS in Data Management & Analytics, 2024) to measure what I had been supporting. The four repos above are that pivot made public: the same denials, charges, upgrades, and queues, now with labeled ground truth, benchmarks, and CI behind every claim.
The fastest way to evaluate me: git clone any pinned repo and run the three-command quickstart.