APPLIED AI · DILIGENT
Agentic third-party risk intelligence.
I led architecture and delivery from concept through beta, general availability, and enterprise rollout—building retrieval and reasoning workflows for third-party risk and adverse media screening.
~93% lower AI cost$8 → $0.55 per screening
Behind the work +
- The problem
- AI screening had to be commercially viable while giving enterprise customers useful, measurable results. At $8 per screening, the original workflow was too expensive.
- My ownership
- I defined the architecture, built agent workflows and evaluation infrastructure, and provided technical leadership to a 20-member engineering team. Product rollout was a cross-functional team result with Product, Sales, Customer Success, and security partners.
- The engineering decision
- I redesigned agent workflows and prompts, replacing expensive inference with deterministic search where appropriate. An evaluation framework using approximately 5,000 human-annotated records combined precision/recall benchmarks and LLM-as-a-judge evaluation to detect regressions.
- Outcome and scope
- AI cost per screening fell from $8 to $0.55, approximately 93%. The platform deployed across nine AWS regions through one CI/CD pipeline. The cost figure measures AI screening cost; the region count describes deployment scope.