Open source
Kwola
I built an AI system that explores web applications to find bugs. The project combined reinforcement learning with browser automation.
Explore Kwola on GitHubBradley Arsenault · Applied AI Architect
I build AI agent teams that replace a roomful of experts.
I work with specialists to understand how they reason through hard cases: the evidence they weigh, the judgment calls they make, and the exceptions that don’t fit the usual pattern. Then I design and build teams of AI agents that take on that work, each with a clear role, and with clear points for expert review.
See selected workSelected work
The domains change. I keep coming back to problems where software has to make sense of messy real-world inputs and reach a sound conclusion.
Test generation
Checksum.ai
I built models that learn from recorded user sessions and generate tests for how people actually use a product.
Document processing
Documents and receipts
I built machine learning systems that turn documents and receipts into structured information other software can use.
How I work
Work through real cases
I sit with the people doing the work and follow a case from the first input to the final decision.
Map the evidence and judgment calls
I look at what evidence people rely on, what they do when it’s missing, and where a case needs an expert’s judgment.
Build and evaluate the system
I design the models, agents, and integrations, build in expert review where it’s needed, and evaluate on routine cases, ambiguous cases, and failures.
I’m interested in Staff and Principal Applied AI roles where I can own a hard problem from discovery through delivery.
Connect with me on LinkedInHow I got into applied AI
In 2015, I was working on receipt processing at Sensibill. Receipts are easy to recognize and surprisingly hard to turn into useful data. After months of working with traditional machine learning, I tried training my first neural network on the problem.
What stuck with me was what the model made possible for the work itself. I’ve been following that thread ever since, from teaching software to explore web apps for bugs to building systems that work through complex material and surface what an expert needs to review.
