Case Studies
Most of my work lives behind enterprise walls, so instead of repos, here are the stories: the problem, the constraints, the architecture, the decisions, and what happened.
AI Call Analytics: Natural-Language Answers to 'Why Do Customers Call?'
Sole architect of a production AI pipeline ingesting ~1,400 calls/day for a multi-billion-dollar rental company — then used its own data to cut total call volume ~30%.
- ~1,400 calls/day ingested and classified
- Classification: hours → minutes with Batch Inference
- ~30% reduction in total call volume
- First production AI system at the client
Testing Against Production in Seconds: Module-Federation Override Tooling
Built a Webpack 5 runtime-override system and browser extension for a 60-micro-frontend platform where no lower environment matched production — turning hours of setup into seconds.
- 60 micro-frontends, 4 teams
- Prod-parity testing: hours → seconds
- One-click dependency-aware deployments
The $1.4M/Year Bill Nobody Was Watching
A major retailer's RFID loss-prevention POC was drowning in false flags and billing ~$119K/month on GCP. Rebuilt the detection algorithm around windowed differentials, validated the economics with a hand-built usage calculator, and cut the bill ~99%.
- ~$119K/month → ~$1.3K/month (~99% cut)
- ~10 valid events per 10,000 sent, before the fix
- $500–700M ROI opportunity sized in proposal
Being the Entire Product Team Across Two Fashion-Tech Businesses
Recruited to spearhead systems architecture for two partnered companies — a franchise-able bespoke manufacturer and a digital tech-pack startup. Untangled a shared AWS environment, split one overloaded database into three purpose-built systems, and rebuilt the tech-pack product end to end.
- 1 shared database → 3 purpose-built systems (Tech Packs, WMS, WES)
- Manual SSH deploys → CI/CD with health-checked cutover on ECS Fargate
- 95%+ of the tech-pack full-stack app built solo
The Batch Job That Was Silently Corrupting $10M/Day
Three months into the job, a self-directed audit of a 3-year-old batch system uncovered a validation gap letting multi-stop routes ship unvalidated — a $10M/day exposure. Found, confirmed, and fixed the same day.
- $10M+/day in losses prevented
- Silent defect — no errors, no alerts, 3+ years in production
- Found, confirmed, and fixed in one day