Travis Terry
Resume

Staff Software Engineer · Agentic AI Engineering, AI Applications, and Web Platform

Travis Terry

I make AI agents do real, verified engineering work: at Atlas UP my loops took tickets from reproduction to an opened pull request, and billing and auth always stopped for a human.

Download Resume (PDF)EmailLinkedInBook a Call

60-second version ›Sterling Heights, MI (Remote or Metro Detroit)

Travis Terry
  • 0115+years building production web applications
  • 0230+Claude Code skills I wrote as sole author of an agentic engineering system
  • 03aboutabout 10,000monthly active users on the Vue platform I owned

The bug, acceptance, and regression results are in The Pipeline below. Illustrative recreation with synthetic data for all diagrams.

60 sec read

The 60-Second Version

15+ years building production web applications, now focused on making AI agents do real, verified engineering and product work.

01 / 05 · TOP FIVE
30+Claude Code skills, sole author

Sole author of an agentic engineering system at Atlas UP: 30+ Claude Code skills chained into loops that took a ticket from reproduction to a verified, opened pull request.

Stack

MCP server developmentClaude Code skillsCustom commandsAutonomous loopsCursorVerifier agentsLLM code reviewAI-driven test executionMulti-provider model routing
Available nowSterling Heights, MI. Open to remote, or on-site and hybrid in Metro Detroit.
Download Resume (PDF)Book a Call

Atlas UP, June 2025 to October 2026

The Pipeline

At Atlas UP I was the sole author of an agentic engineering system: 30+ Claude Code skills chained into loops that took a Jira ticket from reproduction to a verified, opened pull request.

  1. 1

    Reproduce

    The loop started by reproducing the ticket.

  2. 2

    Fix

    Next, the loop made the fix.

  3. 3

    Verify

    A separate agent, not the one that wrote the fix, verified every fix with fresh context.

    fresh context
  4. 4

    Regression

    The change then ran through regression.

    Related: release regression. At the release level, I built an AI regression lane that ran TestRail cases through Claude Code and Playwright, scoped by each release's blast radius. It cut a full regression cycle from about a week to 8 to 10 hours and removed QA as the release bottleneck.

  5. 5

    Review

    Then came a review step.

    Related: PR review agent. I also built an autonomous pull request review agent on Azure DevOps. It reviewed open PRs at their latest commit and posted severity-ranked findings, encoding team standards so I was no longer the mandatory manual review gate.

  6. 6

    Opened PR

    The loop ended with an opened pull request that included before and after videos.

Ticket 0001 · PR opened

Illustrative recreation with synthetic data

I used the system to clear a backlog of 100+ bugs and about 25 features in about six weeks, with roughly 95% of bug tickets accepted without being bounced back.

The ticket, track, and stations are original drawings. No real tickets, code, prompts, or screenshots appear here.

RELATED: RELEASE REGRESSION

about a week8 to 10 hours

At the release level, I built an AI regression lane that ran TestRail cases through Claude Code and Playwright, scoped by each release's blast radius. It cut a full regression cycle from about a week to 8 to 10 hours and removed QA as the release bottleneck.

8 to 10 hours
Day 1Day 2Day 3Day 4Day 5Day 6Day 7
TestRail casesClaude CodePlaywrightScoped by blast radius
Illustrative scale.
4 safeguards

The Safety Model

I designed the safety model, and every automated action ran inside it.

  1. 01

    Approval gates

    Sensitive areas such as billing and auth always stopped for a human.

    billing / authhuman approval
  2. 02

    Risk-tiered retries

    Retries were capped by risk tier.

    retry cap by risk tier
  3. 03

    Verifier agent

    A separate agent verified every fix with fresh context.

    fresh context
  4. 04

    Global kill switch

    A global kill switch paused all loops.

Try the kill switch

This site has one too. Pause motion stops every animation and smooth scroll and switches to a static layout. Your choice is remembered on this device, and if your system asks for reduced motion, the site starts paused.

bug loop · runningfeature loop · runningregression lane · running

The gate, lanes, and retry meter are original drawings. No real limits, tiers, or settings are shown.

EOS One · production

MCP Server

I was a core contributor on the team that built a production MCP server for EOS One. It let customers connect Claude, ChatGPT, and Grok to their EOS One data.

Routing. A question classifier routed each request to the right tool.

No deletes. Destructive actions, like deletes, were deliberately excluded, so they had to happen in the app, never through an LLM.

TRY A REQUEST
QUESTION“Delete that item”
classifier
TOOLreadsafe
TOOLwritesafe
FENCEDdeleteblocked
Blocked. Deletes happen in the app, never through an LLM.
delete: in the app only. Illustrative recreation with synthetic data.
2006 → 2026

Career Log

Each role is a branch that merged into main. Oldest first, ending at HEAD.

INITIAL COMMIT Education. ITT Technical Institute, Troy, MI. Associate of Applied Science, Computer Science.

  1. June 2006 → March 2011

    SKBK Sotheby's International Realty Birmingham, MI

    Developer and Systems Operations June 2006 to March 2011

    • Maintained the public website and MLS listing feeds (IDX and RETS), and built agent-facing intranet tools in PHP and MySQL.
    • Handled SEO, analytics, and all company IT assets including Windows Server, Exchange, and network hardware.
  2. April 2011 → February 2023

    Qualfon (formerly Dialog Direct and Budco) Highland Park, MI

    Senior Developer April 2011 to February 2023

    Progressed from Email Marketing Coordinator to WordPress Developer, NetSuite Developer, NetSuite Team Lead, and Senior Developer.

    • Built and maintained roughly 15 NetSuite and SuiteCommerce B2B instances serving an estimated 100,000 users.
    • Built a multi-tenant Vue and C# ecommerce platform that deployed customer-specific instances from a shared core, and led its frontend migration from Vue 2 to Vue 3.
    • Led the NetSuite team with five direct reports, owning delivery, technical direction, and client-facing requirements work.
    • Wrote data integrations connecting on-premise Oracle databases to NetSuite, Salesforce, and internal APIs using Jitterbit.
    • Built and analyzed performance and load tests across web and mobile applications, tracing bottlenecks from IIS servers down to slow SQL.
  3. Feb 2023 → June 2025

    EOS Worldwide Remote

    Staff Frontend Software Engineer, Lead Dec 2024 to June 2025

    Senior Frontend Developer Feb 2023 to Dec 2024

    • Built the EOS One component library from the ground up, establishing the frontend architecture, patterns, and standards the product still runs on.
    • Promoted to Staff Frontend Software Engineer, Lead; led three direct reports with one on ones and growth planning alongside hands-on delivery.
    • Expanded from frontend into full-stack delivery as AI tooling matured, contributing across the C# API surface and Python services instead of handing work off.
    • Built AI-powered product features, including voice-activated note-taking and automation inside the meetings platform; added Vitest unit testing and CI/CD quality checks.
  4. June 2025 → October 2026

    Atlas UP Inc. Remote

    Staff Software Engineer June 2025 to October 2026

    Joined through Atlas UP's acquisition of EOS One from EOS Worldwide, with continuous ownership of the product since February 2023.

    • Sole author of a 30+ skill Claude Code system that took tickets from reproduction to a verified, opened pull request.
    • Designed the safety model: verifier agent, risk-tiered retries, human approval for billing and auth, and a global kill switch.
    • Core contributor on the team that built EOS One's production MCP server.
    • Owned the Vue frontend platform for EOS One, about 10,000 monthly active users.
EOS One · Atlas UP

Platform and Leadership

I owned the Vue frontend platform for EOS One, an AI business operating system with about 10,000 monthly active users.

about10,000monthly active users on EOS One
150+components in the library I builtroughly 80% moved to design tokens
35→9retained as one of three staff engineers
5 platform notes
  • 01OwnershipI owned the Vue frontend platform for EOS One, an AI business operating system with about 10,000 monthly active users.
  • 02StackThe platform ran on Vue, C# and .NET, SignalR, Redis, Cosmos DB, and Python microservices.
  • 03Monorepo and tokensI consolidated it into a single npm monorepo and maintained a 150+ component library, which I had built from the ground up, and moved roughly 80% of it to design tokens.
  • 04Component libraryEarlier, at EOS Worldwide, I built the EOS One component library from the ground up, establishing the frontend architecture, patterns, and standards the product still runs on.
  • 05ShippingAt Atlas UP we ran a weekly release train and used Ship/Show/Ask, so changes reached production almost daily on Azure DevOps Pipelines.
  • Code review: At EOS Worldwide, as the sole senior engineer on a predominantly junior frontend team, I personally reviewed every frontend pull request and used review as the main lever for raising the team's baseline.
  • Direct reports: After my promotion to Staff Frontend Software Engineer, Lead, I led three direct reports with one on ones and growth planning alongside hands-on delivery.
  • Mentorship: I have a track record of mentoring engineers into independent contributors.
  • AI adoption: At Atlas UP, I led the team's AI adoption from Cursor to Claude Code to autonomous loops. The rest of the team adopted the skills after seeing the delivery rate.
  • Standards as code: At Atlas UP, I encoded team standards into the PR review agent described in The Pipeline.
  • Retention: I was retained as one of three staff engineers when the post-acquisition team went from 35 people to 9.

Component tiles and team markers are abstract shapes, not real components or people.

2 products, built solo

Independent Work

I built two AI products on my own: FoundrKids, an AI maker-learning platform in pre-launch, and Pocket Chronicle, a native iOS app I shipped to the App Store and later retired.

BRANCH: OPEN · PRE-LAUNCHFounder and Sole Engineer · February 2026 to Present

FoundrKids

In pre-launch

An AI maker-learning platform for kids 7 to 15: a chat coach guides build missions while parents track safety.

140+API endpoints
about 2,500tests
  • I built the full stack solo: three Vue 3 apps, a FastAPI backend with 140+ endpoints, Postgres with pgvector, Redis, and Stripe on Fly.io and Neon, plus an adaptive learning engine.
  • Every model call goes through a six-layer safety pipeline: blocklist, OpenAI moderation, prompt scaffolding, the model, a streaming sentence buffer, and a PII-redacting output filter. It fails closed; critical events end the session and alert the parent.
  • Cost-controlled model routing sends work to Anthropic or OpenAI by mission tier, with cross-vendor failover and a monthly spend cap per household.
  • COPPA is built into the data layer: verified parental consent, a 90-day chat purge, real child deletion, and a tamper-proof audit log.
BRANCH: RETIREDNative iOS app · 2026

Pocket Chronicle

Shipped to the App Store in 2026, now retired. It worked end to end but did not find users, so I shut it down rather than keep paying to run the backend.

A family memory app I shipped solo, from concept to the App Store in about a month, by directing AI agents outside my primary stack.

about 2,700automated tests
200+merged PRs
  • Stack: SwiftUI and SwiftData with offline-first sync, StoreKit 2, Sign in with Apple, and on-device transcription and Foundation Models; FastAPI and async SQLAlchemy on Postgres, deployed to Fly.io.
  • Quality: about 2,700 automated tests, 200+ merged PRs, and a written spec and decision log behind every change.
Claude · grounded in my resume

Ask My Resume

Ask anything about my work. Typed questions are answered by Claude using only my resume, this site, and my own interview notes, and every answer shows its sources. If it does not know, it says so.

The idea comes from the question classifier in the EOS One MCP server I helped build, which routed each request to the right tool.

Pick a question below, or type your own.

Rate limited and checked for bots by Cloudflare Turnstile. If the AI is unavailable, my approved answers fill in.

Ready for review.

If you're hiring for agentic AI engineering, AI applications, or web platform work, I'd like to talk. I work remotely, or on-site and hybrid in Metro Detroit.

EmailLinkedInBook a CallDownload Resume (PDF)
github.com/travist6983

Email me