Work

Four case studies

The first two are from employers, PayPal and Amazon. They use only the figures on my resume and describe design decisions. The last two are my own, so the numbers go deeper.

PayPal 2022 to 2026

Proving every card transaction is accounted for

Technical reconciliation for Lynx, PayPal's in-house card processing platform. An hourly job on the happy path, and an event-driven rerun for any hour that receives late records.

4 teams upstream, across three time zones, none reporting to me
1 code path a late file reruns the whole hour, so first pass and rerun share the code

Amazon 2017 to 2020

Replacing a live telemetry pipeline without breaking the dashboards

I owned the telemetry pipeline and client library for the delivery associate app while its fleet grew from 70K+ associates to about 1M registered. I re-architected both, and moved live traffic with old and new running in parallel.

~1.1 TB / day into a Redshift store of more than 100 TB
70K to ~1M registered associates in the fleet the pipeline had to keep up with

nextbookinorder.com 2026 to now

A bug that ships once gets a deploy check

A reading-order catalogue for book series, more than 68,000 pages, that I built and run end to end. Its LLM pipeline writes nothing when a page has no usable source, and my rule is that each class of bug that ships gets a check in the deploy gate.

18% to 3.3% fabrication in small samples: thin sources first, then a fresh sample after stricter gates
29 to 63 check functions in the deploy gate, Aug 1 to Sep 30, 2026

Engineering practice 2026

Running AI coding agents under independent review

I direct AI coding agents. I give them a written bar and a separate reviewer, and I'm trying out a local model for the mechanical work. Most of my code is written this way.

116 of 133 must-fix findings on 19 mostly-documentation pull requests, confirmed by independent refuters
5 of 6 verifier agents caught reporting clean without running their assigned check

Open to Staff and Lead backend roles

Remote in the US, or hybrid in Austin, Texas. LinkedIn is the fastest way to reach me.