Work
Case studies of systems I designed and shipped at Chipper Cash: the problem each one solved, the architecture behind it, and what changed once it was running in production.
Series · 3 case studies
An LLM system for sanctions screening, in three parts: the architecture, the evals that earned its trust, and what production taught us.
A two-model pipeline that adjudicates sanctions-screening alerts: it clears the routine false positives and escalates anything that looks like a real hit.
A golden dataset built by double-blind review, a confusion matrix, and the one recall number we refused to trade.
Draining a backlog of ~148K screening hits at four cents each, and the day the pipeline reported success while doing nothing.
Series · 3 case studies
The recurring shapes of consumer-fintech fraud: a field guide to the archetypes, a deep dive on the one that hid inside real growth, and the automation that runs the fraud desk.
A primer on the fraud archetypes of consumer fintech: how they work, the tells that give them away, and how to stop them.
How we separated genuine super-referrers from referral fraud.
An LLM-powered pipeline that reads partner fraud reports, resolves identity, and acts in seconds, replacing 30 minutes of analyst transcription with 20 seconds of automation, 24/7.
Series · 3 case studies
Replacing a $250K/year vendor with an in-house platform: the decision, the architecture, and the detection rules that power it.
Why I replaced a $250K/year third-party case-management vendor with an in-house platform. Two years on, it has carried 53,000+ alerts, 20,000+ cases, and 900+ regulatory filings across seven markets.
The technical architecture behind the in-house case-management platform: the transaction-monitoring engine, the six-table data model, and the controller layer.
How I standardized 35 transaction-monitoring rules across seven markets using dbt, and the discipline of tuning, rerouting, and retiring rules in production.
Case study
The dbt runtime and the data layer everything else is built on.