Data Foundations & Governance
Lineage, quality, and governance built to carry AI workloads — the layer most AI programs discover too late.
The approach
Most AI initiatives do not fail on the model. They fail on the data underneath it.
A pilot works because someone hand-picked the inputs. Production fails because nobody can trace where the data came from, nobody measured its quality, and nobody governed who could reach it. By the time this surfaces, the model has already been blamed.
This is the work I specialize in. Data lineage you can actually follow, quality instrumented rather than assumed, and governance that survives an audit — designed specifically for the demands AI places on a data estate, which are different from and heavier than traditional analytics.
It applies whether you are preparing for your first AI use case or trying to work out why the last three stalled.
Proof
Retiring a Paid Data Vendor
A provenance and confidence layer over 7.4M healthcare providers — re-sourced onto authoritative registries, with every field carrying where it came from and how much to trust it
- 7.4M
- Providers in the enriched spine
- 94.9%
- License records from authoritative sources
- 34
- State licensing boards onboarded
- 14
- States with paid verification fully retired
Documentation That Cannot Go Stale
A governance platform that reverse-engineers a 322-table healthcare data model into published, business-readable documentation — regenerated from live sources every morning
- 2,451
- Fields documented and curated
- 322
- Tables covered
- 213
- Invisible data gaps surfaced
- 65+
- Consecutive unattended daily runs
Governance Out of Spreadsheets
A spec-driven admin application that let business analysts own a healthcare provider data model directly — versioned, auditable, and with no engineer in the loop
- ~4 weeks
- Build time, solo
- ~2,100 lines
- First-party source
- ~10 lines
- Cost of adding a governed table
- 0
- Engineers needed for a mapping change
Moving a Catalog That Cannot Be Moved
A recreate → deep clone → cutover playbook for migrating Unity Catalog storage roots, with a mandatory human gate in front of the one irreversible step
- None exists
- SQL command that moves a catalog
- 5 of 6
- Reversible steps in the playbook
- 1
- Mandatory human sign-off gates
- Yes
- Delta history preserved
Owning the Risk, Not the System
Setting target-state architecture and holding approval authority on the client side, as a national healthcare registry moved off a vendor-controlled stack it could not see into
- 250M
- Member records ingested weekly
- 1.2 TB/wk
- Compressed volume at the raw layer
- 12 weeks
- Architecture & initiation phase
- 6
- Medallion layers designed
Other services
AI Strategy & Roadmapping
A sequenced roadmap with named owners and honest estimates — not a slide deck that sits in a drawer.
View approachAI Governance & Compliance
Governance mapped to NIST AI RMF, ISO 42001, and the EU AI Act — written by someone who has built the systems being governed.
View approachAI Platform & Application Development
Production-ready AI systems that integrate with your existing infrastructure — evaluated before launch, observable after it.
View approachAI-Augmented Engineering
AI pointed at your delivery process — feature decomposition, commit-level quality gates, and documentation that stays current.
View approachCourse & Curriculum Development
Technical curricula built as engineering systems, so a course can be corrected in an afternoon instead of rebuilt every few semesters.
View approach