Thomas Hsieh

Thomas Hsieh

AI enablement leader.

I turn noise into signal.

I lead AI rollout for program finance and scheduling in aerospace and defense, from the roadmap and governance to 6 RAG agents released to 40 analysts and schedulers.

What I do

I help teams adopt AI in the work they already do. AI enablement is a business process change before it is a technology one. I start with what people actually do and how they behave, design the experience around that, and then deploy AI where it turns noise into signal.

Eleven years in cost, schedule, process improvement and reporting systems taught me where the noise comes from.

How I run AI enablement

  1. Prioritize use cases with the people who own the work, starting from decisions they make every week.
  2. Have the owners document the work, so the AI answers from what they wrote.
  3. Put the rules in writing: what AI can find, compare, draft and route, and where a qualified person has to review.
  4. Train people by role, with a champion in each team who answers questions after training.
  5. Set targets with the business owner before launch: repeat use and one task result after a month. Expand only where both are met.

What I've built.

AI rollout for program finance and scheduling

Aerospace and defense · General Atomics · 2018–now

Since 2025 I've led AI rollout for program finance and scheduling teams. I presented its roadmap and strategy to group leadership, and wrote the rules for where a qualified person reviews AI work.

  • 6 RAG agents released to 40 analysts and schedulers
  • Governance framework for who approves data access and dashboard releases
  • Tools and reporting that reach 200+ program staff
  • Trained 40 to 50 cost and schedule analysts
  • Three company awards for tools and process work

Cortex, my governed agent system

Personal project · built with Claude Code and Codex · 2026

It reads my mail and messages, keeps a written record of my life and commitments, writes my morning brief, and runs always‑on assistants I talk to by chat.

I built it the way I'd want AI rolled out at a company. Every capability has a written permission sheet, safety rules are enforced in code, and no AI decision goes live until I've measured how often it is wrong.

240+ written specs so far, each reviewed by a second AI model before it's built.

On a 200‑email test, its classifier sorted 82% on its own at 96% accuracy. Plain rules handle the rest.

Halfdoor, one shared memory for your household

Founder · 2025–now · in private testing

It gives a household one shared memory inside Claude and ChatGPT. Nothing is shared unless someone saves it to Home.

  • Works in Claude and ChatGPT as a connector (MCP)
  • A private memory for each person, and a shared one for the household
  • Earlier version: agentic search over 13,000+ appliance manuals, with the page cited

VoiceFlow, a voice‑to‑text keyboard

Open source · Android

Record, transcribe, clean it up with your own AI prompt, and insert it into any app. Bring your own key, or run fully offline on the phone.

Built because I didn't want to pay $30 a month for a dictation app.

See it on GitHub

How I work

I view AI as a way to reclaim our lives.

A serial creator.

Self-taught in animation, music production, iOS development and UI design, because I'd rather learn the tool than wait for someone else to. These days I build mostly by directing Claude Code and Codex.

From program controls to AI, by way of building things.

  1. 2025–nowHalfdoorFounder
  2. 2018–nowGeneral AtomicsProject Control Analyst, then Business Process Analyst (AI Enablement)
  3. 2022AntlerFounder in Residence
  4. 2020–2024Side venturesFounded Soulbii, a virtual business card built with a three‑person dev team. Ran operations for NeedThat. Built Asian Hustle Network's website, member directory and marketplace.
  5. 2015–2018Northrop GrummanProgram Control Analyst
  6. 2010–2013University of California, IrvineB.A. Business Economics

Have a noisy problem? Let's find the signal.

If you're building an AI enablement function and want people to actually use what it ships, I'd like to hear about it. Based in Austin; open to remote or hybrid.

© 2026 Thomas Hsieh