All workFILE 01 / 04

Banking

Fraud decisioning at scale

Shipped at Citibank · led by Shashwat Verma

01

The problem

16 million card transactions a month need a fraud call in milliseconds. Block too much and you lose customers; too little and you eat the loss.

02

The constraint

Bank-grade model governance: every model validated (KS, PSI), fair-lending reviewed, and auditable.

03

The system

Fraud decisioning models for Best Buy, LL Bean, and Home Depot portfolios, with policy thresholds tuned per portfolio and a validation trail regulators accept.

FIG. 01 · SYSTEM FLOW · 7 NODES · 7 FLOWS

Card transaction stream → Neural fraud model (score). Bureau triggers → Per-portfolio policy thresholds. Neural fraud model → Per-portfolio policy thresholds. Per-portfolio policy thresholds → Approve · decline · review. Approve · decline · review → Risk analyst review (edge cases). Risk analyst review → Validation & audit trail (labeled calls). Per-portfolio policy thresholds → Validation & audit trail (KS · PSI)

04

The human in the loop

Edge cases route to analysts with full context; the system learns from their calls.

05

The outcome

16M decisions/month, 5+ years in production.

← FILE 04 · AD PLATFORM (IN-HOUSE)AI video ad production pipelineFILE 02 · DRIP CAPITAL →Risk portfolio & transaction policy
Start your own file