Career timeline / what I built, in order

The builds, in order.

A dated record of what I have built, from my first Python scripts in 2022 to the software I ship now. Each date comes from a file, a commit, or a published page.

Entries
46
Animated scenes
31
Span
2022 to now

Scroll, or wait a moment. It plays by itself.

Client names are left out. This is selected, not exhaustive.

2022

Learning to build, in public

I left insurance sales and started to write code. Within a year I had worked through a deep learning textbook, published a children's book made with AI, and shipped my first hardware builds.

  1. Start

    First Python scripts

    Practice programs and Raspberry Pi camera timelapse scripts. This is the earliest code of mine that I still have.

    Dated by File dates

  2. Milestone

    Left insurance sales after seven years

    I had been a top producer and had led a field sales team of 22 agents.

    Dated by Résumé

  3. Study

    Deep Learning with Python, chapter by chapter

    I worked through the book in more than twenty Colab notebooks: a linear classifier from scratch, movie-review and housing-price models, overfitting and validation, the Keras functional API, a custom training loop, and a convolutional network that tells cats from dogs.

    Dated by Notebook dates

  4. Field work

    First weather-balloon tracking station

    I started to log radiosonde telemetry with a home receiver. This hobby later became Sondefinder, Cold Hunts, and a magazine.

    Dated by Receiver logs

  5. Experiment

    First images from DALL-E 2

    Beta access arrived. I spent hours finding out what a text prompt could and could not do.

    Dated by Image file dates

  6. Experiment

    First code written with ChatGPT

    Within a day of ChatGPT's public release I used it to write MicroPython for a model car's lights. My notebooks from the same week hold a weather display, a flight tracker that sends texts, and a small CRM.

    Dated by Notebook dates

2023

Teaching AI, then shipping software

I taught what I had learned: a webinar, a podcast, and a course for homeschool students. Then I shipped my first web apps.

  1. Course

    Future Homeschool: Intro to AI

    An eight-week course for homeschool students, with an exercise each week.

    Dated by Course records

  2. Release

    Color Palette Tester

    My first shipped web app, and my first product built on GPT-4. It was the first of four design tools I built for The Color Palette Studio.

    Dated by Git history

  3. Release

    Sondefinder

    A map and search tool that finds weather balloons that landed near you. It is still online.

    Dated by Git history

2024 – mid 2025

Design tools for working designers

For The Color Palette Studio I built tools that graphic designers use to choose and test brand colors. Between releases I built small aviation devices.

  1. Release

    Brand Visualizer

    Shows a brand's fonts and colors on realistic mockups, so a designer can present a direction and not a swatch sheet.

    Dated by Git history

Jul 2025 – Mar 2026

AgentWorks: AI and integrations for service businesses

I co-founded AgentWorks and ran its engineering. We set up CRMs for home-services companies, and I built the software that connected those CRMs to the rest of each business. On my own time I went back to aviation.

  1. Client work

    Contact research agent

    A new lead arrives in the CRM. An AI agent researches the person and the company on the web, then writes a cited briefing into the lead's record before the first call.

    Dated by Git history

  2. Client work

    Two-way sync between a CRM and a job-management system

    For an artificial-turf installer. Leads moved one way in real time, and proposals and approved jobs came back each night. The system kept an audit log and could find and repair records that did not match. This was the largest build of the year.

    Dated by Git history

  3. Client work

    Lead and quote forms for eight companies

    One hardened form system, fitted to each business: two pool builders, a turf installer, a remodeler, a garage door company, a patio builder, an insulation franchise, and a gutter company.

    Dated by Git history

  4. Client work

    Signed estimate to crew work order

    A tool that takes a signed estimate, removes the prices and the terms, and makes the work order that the crew needs.

    Dated by Git history

  5. Client work

    Lead routing and franchise-software sync

    One service matched each new lead's address against past jobs and assigned the correct sales rep. Another moved leads out of software that a franchise required and into the CRM every five minutes.

    Dated by Git history

  6. Research

    Detect-and-avoid, trained from camera pixels

    A GPU-parallel reinforcement-learning environment for a fixed-wing aircraft that must see and avoid other traffic with a camera alone. I trained the first models in it.

    Dated by Git history

  7. Client work

    An orchestration brain for the CRM

    The move from single tools to a platform. AI proposes the next step with a lead, and fixed rules decide if it can happen: quiet hours, do-not-contact lists, and a human takeover. Every action is recorded. It ran first in a shadow mode for a specialty retailer. Then I designed the next version, in which each client has an isolated copy.

    Dated by Git history

Mar 2026 – now

Lightcone Systems and Ledger

I founded Lightcone Systems, built Ledger, an operational record for drone programs, and released it as open source. The same year I shipped weather tools, and I now do integration work for businesses in Nashville.

  1. Essay

    Part 108 Is Coming

    Why American drone founders need tools, and not only rules, when flights beyond visual line of sight open up.

    Dated by Published post

  2. Talk

    AUVSI XPONENTIAL, Detroit

    I presented at AUVSI XPONENTIAL 2026.

    Dated by Public posts

  3. Release

    Ledger public beta

    The second version opened to operators, with rule packs for the FAA's drone regulations and a server that lets AI agents read the record.

    Dated by Public posts

  4. Engineering

    Four more clean-sheet rebuilds

    Each time I learned what the record had to guarantee, I started again from an empty repository. Ledger became a records platform with a history that cannot be edited, then a safety management system built on that record. Six versions in eight months.

    Dated by Git history

  5. Open source

    Ledger released as open source

    The operational record is now open for anyone to read, run, and check. It includes rule packs for the FAA's Part 107 and the proposed Part 108. People make each finding, and Ledger keeps the record.

    Dated by Git history

What comes next

The next entry could be your workflow.

I connect the software that a team already uses, and I start with the one workflow that costs the most time.

Browse the public record

Now showing Apr 2022