Case Studies
Real projects. Real results. Evidence, not marketing.
Detecting $500M in Medicare Fraud with Public Data
A single Go binary ingests 548 million rows from 22 CMS datasets, scores 1.47 million Medicare providers, and serves an interactive investigation dashboard. 42% Precision@100 — an 890x improvement over random. Built in 13 days by 4 people. Infrastructure: under $800/month.
4M Lines of Java/XML to Production TypeScript in 75 Days
A mission-critical federal integration platform built on Java and a custom proprietary XML language. Previous modernization attempts had failed. A 2.3-person team replaced it with 120K lines of modern TypeScript — 85% AI-generated code with full test coverage. 7 bugs at launch, 1 high-impact.
Replacing a 100GB Graph Database with a Function Call
40+ graph algorithms for Go. Built to replace Memgraph — eliminated 100GB RAM, Docker dependency, and dual-license risk. Pure Go, 97.3% test coverage, MIT licensed.
12 Million Words, 2 Days, 1 Engineer
AI-powered extraction pipeline processed one of the longest works of fiction ever written — 12 million words across 807 chapters — into a searchable interactive map. 100% AI-generated code.