Hi, I’m Jason.
I run a studio of one, designing and operating AI products end to end. I prefer the quiet kind — products that do real work for real people, not demos that vanish after the screenshot.
I came to this work the long way around: a PhD in AI, then more than ten years across research, engineering, and the unglamorous middle where ideas have to become systems that hold up under real workloads. I’m at home across the layers — the mathematics underneath, the model architectures, the tooling, the operational realities that decide whether a system survives outside the lab. To me these aren’t separate disciplines but the same problem at different magnifications.
Right now my main project is Clearfraim, an AI data and BI platform for small teams. Alongside it, I build 67Quant for trading, IMPG for creator brands, and Saja for voice agent automation.
I write here about what each of them teaches me — about designing agents, evaluating them, keeping them from drifting, and running my own marketing on the same patterns I sell.
AI advice is everywhere.
AI that ships is rare.
The market is saturated with strategists, frameworks, and demos that vanish after the screenshot. Software that survives real users, real models, and real time is harder to come by. That gap is where the studio sits — closer to a working operator than a consultant with slides.
One operator, end‑to‑end.
Architecture, build, evals, and ops — same person, same brain. A studio of one is a feature, not a bug: no specialist handoffs, no slideware between idea and production, no finger‑pointing when something drifts at 2 AM. Everything that ships from the studio runs on this discipline, including this site.
Software that holds up.
Products that survive real users. Automations whose evals catch drift before customers do. Engagements where the lessons compound — into the next product, the next field note, the next client. The writing, the products, and the work all feed each other. That’s how a studio of one keeps up.
Four threads, one studio.
Agent design
Multi-agent workflows that survive contact with real users — evaluations, tool use, memory, and human-in-the-loop where it matters most.
Voice & telephony
Voice agents that pick up the phone and finish the job, with elegant handoff to humans for the long tail of edge cases.
Data into decisions
Turning operational data from siloed tools into something a small team can actually act on — not another dashboard nobody opens.
Growth, automated
Replacing the manual marketing-ops tax — content, posting, analytics, email, CRM — with a quiet, coordinated agent fleet.
“The best AI products feel less like software and more like having a good colleague who happens to never sleep.”
A small number of engagements each quarter.
Usually advisory or contract work — helping a team design their first production agent, untangle an eval setup, or ship a voice or data product. If that sounds like the right fit, say hello.