Taylor Perry
I build backend systems where correctness matters, and AI tools that help engineering teams understand those systems.
Austin, Texas
I build backend systems where correctness matters, and AI tools that help engineering teams understand those systems.
I'm a senior software engineer at Shippo, where I've worked since 2021 on the systems that move the money behind shipping labels: buying them, auditing what carriers charge for them, and automating the decisions in between. Most of that work is backend: event-driven pipelines, workflows that have to survive retries and restarts, and the infrastructure and observability that keep them correct in production.
More recently that work has grown into applied AI. As part of Shippo's engineering-wide AI effort, I build agents that investigate these systems and the knowledge infrastructure that keeps documentation grounded in the code that actually shipped.
API
The first product in Shippo's automations initiative. A merchant defines their shipping strategy once, as rules; one API call picks the right label, buys it, and returns a decision trace explaining why. I'm one of two engineers building it: the rules engine, the Postgres-backed workflow that can die mid-purchase and recover without buying a second label, and the end-to-end gate that blocks a deploy unless the telemetry it expects actually shows up.
Pipeline
Carriers bill Shippo for every label it sells, and no two agree on how: EDI, CSVs, password-protected spreadsheets, PDFs, over SFTP, APIs, and email. Five services turn that mess into Shippo's source of truth for carrier cost, every charge matched to the label it belongs to. The platform reconciles hundreds of millions of dollars of invoices a year, recovers tens of millions, and audits over a billion in carrier withdrawals. I was a founding engineer, from the first commit: most of the carrier normalizers, the Polars matching engine, the merchant billing stages, and the SLO model the team still operates against.
Agent
Billing investigations used to mean an engineer spelunking through Databricks and logs. This Slack agent does the spelunking: ask in a thread, and a sandboxed analyst subagent queries Databricks and New Relic, then streams back evidence-cited findings. Read-only access is enforced in code, not by a prompt. Its knowledge base is a versioned pack teammates extend through pull requests and load into their own agent sessions; Shippo's agent platform docs cite it as the working example for internal operational agents.
Tooling
Documentation drifts because keeping it honest is bookkeeping nobody has time for. This harness makes the bookkeeping autonomous: merged PRs, design docs, transcripts, and working sessions compile into an in-repo knowledge base holding only what the code can't show (rationale, cross- system contracts, invariants, open questions), every claim checked against merged code first. Deterministic guardrails enforce the format; the model does the judgment. It runs in CI at Shippo, feeding the org's AI code reviewer and a cross-repo knowledge server.
Infra
The AWS infrastructure for four production services on EKS, built and operated as code: IAM and OIDC, databases, messaging, secrets, Helm deployments through Argo CD, canary rollouts with Argo Rollouts, and the New Relic dashboards and alerts that watch all of it.
2021–present
Shipping automation, carrier reconciliation, and applied AI tooling, described above. Promoted three times.
2021
Designed and implemented the Django REST API for a public astronomy database of infrared spectra from planet-forming disks.
2020
Built and ran the first website for a Texas history brand with more than a million Facebook followers, tying together its community forum, photo storefront, supporter platform, and social channels.