Axon

The Context

Axon builds technology for public safety. Their mission is to Protect Life and Obsolete the Bullet. When I joined in July 2022, the AI team was shipping computer vision, speech recognition, and language model products, but the responsible AI practice was still early. There was no systematic way to evaluate models for fairness, no automated pipeline for ethical assessments, and no platform that could scale these practices across use cases.

That was the gap I was hired to fill.

What I Built

The Responsible Innovation Platform

Over two years, I built and led the team that created Axon's responsible innovation platform. Every AI use case, whether it was computer vision, LLMs, or ASR, had to pass through this platform before reaching production. The goal was to make responsible AI the path of least resistance for engineering teams, not a compliance gate that slowed them down.

I led a team of machine learning engineers who built the tooling, the evaluation pipelines, and the review processes. By the time I left, the platform was a core part of how Axon shipped AI.

International ASR Expansion

Axon's Automatic Speech Recognition products needed to expand internationally. That meant evaluating model performance across new languages and dialects, and doing it in a way that was fair and unbiased.

I built an automated evaluation workflow that could assess the ethical performance of ASR models across different demographics and locales. In 2023, we used it to evaluate models on 4 new locales.

Privacy-Preserving ALPR Evaluation

The Automatic License Plate Recognition product had a similar challenge. We needed to rigorously evaluate model performance without ever seeing the underlying data, because the data was subject to GDPR and other privacy regulations.

I designed and built an automated, privacy-preserving ALPR evaluation workflow that operated in an eyes-off, GDPR-compliant environment. This proved that you can do serious model evaluation without compromising data privacy.

AI Platform (IC Phase)

Before taking on the management role, I spent my first year as a senior IC leading a team of 4 Machine Learning Engineers. We worked on ALPR, ASR, and LLM applications that were used by 5+ research scientists. This was where I developed the technical foundation that informed the platform I would later build.

The Arc

I joined as a Senior Machine Learning Engineer II, was promoted to Tech Lead after six months, and became Manager of the Responsible Innovation Platform a year after that. The progression reflected the growing importance of responsible AI at Axon and the team's expanding scope.

What I Learned

Building responsible AI inside a company whose products are used in law enforcement taught me that this work is not abstract. The stakes are real. Biased speech recognition can affect the outcome of an investigation. Unfair license plate recognition can lead to wrongful stops. The evaluation pipelines I built were not academic exercises. They were safeguards.

I also learned that the hardest part of responsible AI is not the technology. It is getting engineering teams to adopt the practices voluntarily. That requires making the tools good enough that people want to use them.