Featured · Portfolio
Build by Build
Eleven builds. Fifteen test surfaces. Evidence with its limits kept in view.
AI Analytics Architect
12+ years turning messy data into decisions leaders trust.
Open to analytics manager, lead, or principal roles where reporting has outgrown trust, and to conversations about roles that don’t exist yet. Seattle or remote.
The one rule
“Validation suites that fail the build rather than ship a number I can’t defend.”
Career figures from the About page; portfolio figures from the Build by Build inventory, frozen 2026-10-07 baseline; additive extension 2026-10-19.
Case studies
Cascadia is a portfolio built to be checked, not taken on faith. Which tests each module could carry, build by build, is one page: Build by Build.
Featured · Portfolio
Eleven builds. Fifteen test surfaces. Evidence with its limits kept in view.
11 builds × 15 test surfaces
Solid: recorded · Split: limited
Dim: not carried · Dashed: not assessed
Hollow: not applicable
Last row: Claim Integrity. Artifacts are not pass ratings.
03Playbook
Analytics Validation / Playbook
How can you be confident in code you did not read or write in full? One invented dispatch rule, a known answer specified before the code, and four experiments that run a real SQL query in your browser and show what a check adds to reading.
02Playbook
BI Leadership / Operating Model
How I would run a data, analytics and automation team inside a medical device quality organization: principles set in stone, methods adjustable, and every practice carrying a proof, a test and a mark. A scenario, not a record.
12Playbook
BI Leadership / Metric Governance
A 24-week program design for moving an organization off legacy BI and onto a governed, AI-assisted platform, with a certified metric layer at the center. A design, not a build.
04Case study
Healthcare / Claim Evidence
Trace supported claim facts into target readiness and encounter outcomes. All 98 applicable fixture variants pass; 22,173 numerical cells match. Chart review and a four-seat restricted simulated panel are complete. Frozen synthetic demonstration, with scope and independence limits disclosed.
05Case study
Medical Device Quality Analytics
A forecast of FDA device reports three months ahead for one cardiovascular product code, with an explicit range that held in 20 of 22 locked-test months. Counts of reports, never rates, and no firm named anywhere. Public openFDA data.
06Case study
Revenue Operations Analytics
What a customer is contracted for and what they are billable for are different numbers. Three asymmetric timing rules keep them apart for up to a year with nothing erroring. Derived twice, by two engines written to be different, agreeing on every cell. Synthetic data throughout.
Method
01Specify
Validation PlaybookSpecification → agent build → review
I set the architecture, the data model and the measurement definitions as a written specification. AI coding agents build to it. I do not hand-write the Python: I write the spec, review the output and validate it.
02Validate
Revenue AssuranceDefinition → independent SQL → reconcile
A separately written calculation checks a published figure against its definition. Revenue Assurance derives every published cell twice, by two engines written to be different, and they agree. Claim Integrity also publishes its numerical agreement and the separate limits of its assurance record.
03Operate
Matter LedgerSchedule → re-assert → publish failures
A surface that has only ever shown green has demonstrated nothing. Matter Ledger runs on a schedule, re-asserts its own invariants on every run, reconciles a live increment against the frozen baseline, and publishes the result, including when a run fails.
17Process design
Process design · A working system
A job search mapped as a working process: Stage-Gate for structure, BPMN swimlanes for who does what, and a plan-do-check-act loop inside every interview round. Two phases, five lanes, and two keys at every interview gate.
18Paper
Mathematics · A pet project
Not an analytics module: a short mathematics paper. The number of ways n children can take sides against each other, with the rest uninvolved, is (3n − 2·2n + 1)/2, the Stirling number S(n+1, 3). Of the 25 ways four children can, 6 are one against one; of the 3,025 ways eight children can, 28 are.
Let’s talk
Comfortable as the person building it, or as the person who stands up the function and hires the team. The standard doesn’t change either way.