Skip to content
Mitchel Carson

About

Portrait of Mitchel Carson

From Air Force Two to applied AI.

I came to AI through operations first. From 2020 to 2023 I was an Executive Missions Aviator in the U.S. Air Force’s 89th Airlift Wing, flying on more than 50 executive airlift missions across more than 30 countries. Every one was completed with zero safety-related incidents.

Background

Background

I worked directly onboard with distinguished visitors, including the Vice President aboard Air Force Two and the Secretary of State, managing passenger safety and logistics with flight crews, security teams, and White House staff. I also handled the mission administration: in-flight meal and beverage service, baggage, crew hotels and ground transportation, visa applications, and subsistence billing. It was while serving that I chose to pursue AI. I still hold an active TS/SCI clearance.

I studied computer science at Appalachian State, graduating cum laude with a senior honors thesis on runoff forecasting with deep learning. Experiencing Hurricane Helene in Boone is what motivated HYDRA, my research on watershed dynamics and forecast reliability. At USAA I built Java and Spring Boot GraphQL APIs and cut troubleshooting time by 30%, and I also build and advise on software for a small business.

Today I am completing an M.S. in Artificial Intelligence at UT Austin (4.0 GPA), with HYDRA as ongoing research: a results manuscript for Water Resources Research is in preparation, and a software paper for Environmental Modelling & Software is planned. I focus on applied AI that makes complex data useful to the people relying on it.

Service

What I carried into engineering

I hold an active TS/SCI clearance and welcome conversations with defense, intelligence, and other mission-driven AI teams.

  • 01

    Preparation is the job.

    A mission was planned long before anyone boarded, down to the hotels, the visas, and the in-flight service. I build research the same way: data manifests, configuration, and leakage checks settled before a model trains.

  • 02

    Communicate up, down, and across.

    I worked alongside flight crews, security teams, and White House staff. Now it is product owners, engineers, and scientists, and the habit is the same: say what is known, what is not, and what happens next.

  • 03

    The standard does not move.

    More than 50 missions, zero safety-related incidents. I hold my own work to that bar: evaluation that matches how forecasts are really used, and no result published before the analysis supports it.

Full service record

Experience

Selected experience

  • Executive Missions Aviator · 89th Airlift Wing · 2020–2023

    United States Air Force

    I flew on 50+ executive airlift missions across 30+ countries, all completed with zero safety-related incidents, working directly onboard with distinguished visitors including the Vice President aboard Air Force Two and the Secretary of State. I also handled in-flight service, baggage, crew hotels, visas, and billing.

    Ask me about: Operating to a zero-incident standard, executive airlift, and moving from military service into AI.

    Service record

  • Software Engineering Intern · Global Headquarters · 2025

    USAA

    At USAA’s global headquarters in San Antonio, I built Java and Spring Boot GraphQL APIs for customer-data workflows and cut troubleshooting time by 30% with API enhancements and JavaScript comparison views.

    Ask me about: GraphQL API design, customer-data flows, and enterprise delivery.

    View project

  • Software Consultant

    GreenSpace Lawn Care

    I built and launched GreenSpaceLawnCare.us and advise the company on social media strategy, translating business needs into a clear digital experience.

    Ask me about: Client discovery, website delivery, and translating business goals into software.

    Visit website

Values

What I care about

  • Evaluation under forecast-time constraints

    I design models and pipelines teammates can trust, with clear metrics, careful evaluation, and predictable behavior when inputs change.

  • Reproducible by default

    Configs, seeds, data manifests, and tracked runs ship with the result, not after it.

  • Limits made visible

    Good systems make uncertainty, limits, and failure modes visible.

Focus

Focus areas

AreasWatershed forecasting and forecast reliability · Sequence models: LSTM, Transformer, Mamba · Leakage-aware evaluation by lead time · Bounded LLM autonomy in decision systems · Cloud research compute: Google Cloud, AWS · Production software and APIs