Work

Work history

Case studies

Cart.com

2021 — 2026

Director, Technical Product Operations — fulfillment platform · 15-person org, 19 warehouses, ~25,000 shipments/day

Cart.com · Org-wide AI engineering rolloutSpecifics redacted for public sharing

331 → 680 tickets per quarter · flat headcount

Ask
Throughput had plateaued while roadmap pressure hadn't. Two dev teams, a near-shore QA team, a live fulfillment platform — and no new headcount coming. Task: make the org faster without breaking production.
Bet
AI-assisted engineering, org-wide — but demonstrated before mandated. I ran the Claude Code workflow on my own tickets first: 31,808 lines across 264 files in one 18-day window, ~82% fully-achieved task rate, while still running the org.
Catch
A green test suite isn't the same as safe. On one ticket the full suite passed clean — an adversarial diff-read still caught 3 blockers and 6 major issues before merge.
Curveball
Tool mandates die in the backlog. So the workflows became products: a planning skill that investigates the codebase, drafts the engineering plan directly onto the Jira ticket, adversarially reviews its own plan, and hands the ticket over dev-ready. Alongside it, I helped build a spec-based development skill set from the ground up — spec first, implementation against the spec, verification before commit.
Result
Peer-reviewed throughput rose from 331 to 680 tickets per quarter — +105% on flat headcount and unchanged scope. Promoted to Director within the first year post-acquisition.

Snowfall Technologies

2017 — 2021

Development / Product Manager — 3PL fulfillment platform

Snowfall · 3PL platform launchSpecifics redacted for public sharing

0 → 14 clients in 24 months · then acquired

Problem
A newly launched 3PL platform doesn't earn trust on features — it earns it by running other companies' inventory, orders, and shipping without dropping anything. Snowfall had to go from zero clients to a proven operation.
Bet
Build the platform around how a warehouse actually runs — picking, packing, and receiving workflows shaped by an earlier career in fulfillment finance and floor operations, not just software.
Result
14 clients onboarded within 24 months of platform launch — and a platform proven enough that Cart.com acquired it in December 2021.

Engineering projects

Selected work inside those roles

Production AI engineering workflows

Featured

Claude Code · Custom skills · MCP · Jira

What it solves

AI coding tools demo well and then die in real engineering orgs — no connection to the ticket queue, no review rigor, no reason for a skeptical senior to trust the output. These workflows close that gap: they run daily inside a production org, wired into Jira and the codebase.

How I built it

Skills as products
Each workflow (eng-plan, investigate, qa-round, release-validation, track) is treated like a product — iterated against real engineering friction, not demos.
Plans that land on the ticket
The planning skill investigates the codebase, drafts the engineering plan directly onto the Jira ticket, and transitions it to dev-ready.
The plan reviews itself
Before a human sees it, the drafted plan goes through parallel adversarial review — the same skepticism a staff engineer would apply, applied automatically.
Spec before code
I also helped build a spec-based development skill set from the ground up — authored by a teammate, built out together — where a structured spec drives the implementation and the verification that follows.
Demonstrate, then scale
Adoption came from proof, not mandate: I shipped with the workflow myself for a quarter before rolling it out to the org.

Nitty gritty (3 decisions)

01 — Plans live on the Jira ticket, not in a chat log. If the artifact isn't where the team already works, it doesn't exist.

02 — Adversarial review runs in parallel across domains — security, performance, migrations, observability, business logic — because a single reviewer pass anchors on the first problem it finds.

03 — The rollout metric was peer-reviewed tickets, not lines of code — the only number a skeptical engineering org couldn't argue with.

Adversarial multi-domain code review

AI-assisted review · Release quality

Fulfillment platform · Release review patternSpecifics redacted for public sharing

10 fixes · 9 commits · all before merge

Problem
Human review of a large release branch degrades fast — attention is finite and tests only catch what someone thought to test.
Approach
A codified pattern that fans out parallel analyses across five domains — security, performance, migrations, observability, business logic — over the release diff, then merges the findings into one review.
Result
On a single release: 10 reviewer-identified fixes across 9 commits, all pre-merge. On another ticket where the suite passed clean, the diff-read caught 3 blockers and 6 majors.

Orphaned-inventory remediation at scale

Python · Production data integrity

Fulfillment platform · Data-integrity incidentSpecifics redacted for public sharing

4,461 records repaired · zero failures

Problem
A data-integrity defect had quietly stranded thousands of inventory records in a live fulfillment platform moving ~25,000 shipments a day.
Approach
Root-cause first, then remediation built for production: scripted parallel API calls designed for safe scaled execution — idempotent, observable, abortable.
Result
4,461 orphaned records repaired with zero failures, and the root cause closed so the class of defect can't recur.