ChatPrism Labs is an independent research lab in Phoenix, Arizona. We build AI agents for work that is regulated — a patient’s chart, a radiograph, a claim — and we publish the evidence that they hold up.
Mission
Language models are already good enough to take on regulated work. They are not yet good enough to be trusted with it without evidence.
So we build the system and the evidence together. Every agent that leaves this lab carries a hard privacy boundary, a citation for each clinical claim it makes, and an evaluation suite a reviewer outside the lab can run end to end.
That has produced a voice agent that keeps protected health information out of third-party models, a caries detector measured against dentist consensus, and an evolutionary agent search that finds working programs on a fraction of the compute. Next: longer-horizon clinical operations, and the audit tooling that has to come with them.
PHI and PII are detected and stripped before a single token reaches a third-party model, at 180 ms median added latency — with a signed audit trail for every turn.
Read →Lesion-level supervision on 14,300 panoramic X-rays, scored against three-dentist consensus rather than chart labels, and reported per tooth site.
Read →A critic trained on the population’s own failures prunes candidates before execution, cutting the evaluations needed to reach a working program substantially.
Read →The agent assembles the payer packet, cites the chart line behind every assertion, and will not submit without a clinician signature. Nothing it writes is unattributable.
Read →Publications
| Date | Category | Title |
|---|---|---|
| 2026-10-07 | Agentic workflows | An auditable agent for prior authorization, with a human in the loop |
| 2026-09-18 | Voice & Privacy | Streaming PHI and PII redaction for real-time voice agents |
| 2026-08-26 | Evaluation | PrismBench‑Health: a task suite for agents in regulated workflows |
| 2026-07-29 | Medical imaging | Lesion-level supervision for dental decay classification on panoramic radiographs |
| 2026-06-11 | Medical imaging | What image quality costs you: a technical note on panoramic acquisition |
| 2026-05-14 | Agents & Search | Augmenting standard evolutionary agents with a learned critic |
Who we are
Meet the team →We are machine learning researchers and infrastructure engineers who have shipped into clinics and regulated enterprises, working alongside the clinicians who read the images. The lab is small on purpose: the people who do the research also write the evaluations and stand behind the deployment.
Jesse Cisneros
Founder
Rizwan Mehdi
President
Hassan Rehman
AI Solution Architect
How we collaborate
No result leaves this lab on our word alone. Each line of work is reviewed by people who do not report to us.
If you build evaluation harnesses, privacy infrastructure, or medical imaging models, we would like to see your work. If you run a clinic, a payer, or a compliance function and want a pilot that is actually measured, write to us.