Work · evidence
Evidence, not a portfolio.
Three ways to inspect the method: a pro bono client system, two self-built products, and the gaps each one still leaves. No logos, no invented outcomes, no “trusted by” strip.
Index · 2026
AI-Powered BI with Looker
One governed source of truth for a US nonprofit's fundraising operation: four production dashboards, RAG-enriched data, and a LookML layer the volunteer team can own.
Read the case study
4
dashboards live
89→97%
sync accuracy
500+
users
Abyssal Twin
A self-assessment of an open-source digital twin: where production readiness is real, and where it is still a claim.
Read the case study
52%
token spend cut (benchmark)
Real-time
edge inference
Full-fidelity
mission replay
KYC Copilot
My own agentic compliance product—built commercially, then open-sourced: seven invariants, a guardrail before every human review, and a measured 93% reduction in document-review time.
Read the case study
93%
time reduction (benchmark)
14 min
vs 3.5 hr manual
7
enforced invariants
The standard behind the work
Five axes. Twenty-five checks.
Every case study above is read through the same scorecard. See the model, the bands, and what each check asks for.
Own products: Abyssal Twin and KYC Copilot (open-sourced). Pro bono client system: a US 501(c)(3) nonprofit. Client work appears in redacted form; no client logo or named quote is published. Self-assessments do not imply a third-party audit. If a number cannot be reproduced, it does not belong here.