About Meros AI

We build the agents, and we run them.

Meros builds and operates AI agents for specialty medical practices. Not a platform you configure and a licence you renew — we map the workflow that is costing you, build the agent for it, and run it in production behind an evidence trail you can audit.

The problem

Healthcare AI doesn't fail in the demo. It fails at the chart.

Practices don't lack AI vendors. They lack AI that survives contact with the systems they actually run on. Three things kill it, every time.

Failure mode

There is no API for the work

The tasks that eat your staff's day — PDMP checks, pre-visit chart prep, lien packets, note co-signs — live in screens your EHR never exposed to an API. Integration-first vendors quote you a year and deliver a roadmap.

Failure mode

Pilots never reach production

A script that works on a demo account breaks the first time a vendor ships a UI change. Without a way to detect the drift and repair it, the automation quietly stops and nobody finds out until the month closes short.

Failure mode

Nothing is auditable

You are accountable for every line in a patient's chart. "The model did it" is not an answer for a payer, a board, or a plaintiff's attorney. Most AI tooling cannot tell you what it did, when, or on whose authority.

What makes us different

Evidence-grade automation.

Meros is led by two practising physicians alongside an operator and an AI engineering leader: a family medicine physician with three decades of practice and of running clinical organisations as chief executive; a double board-certified, Harvard fellowship-trained interventional pain physician as chief medical officer; and a co-founder who has spent fifteen years between applied research and industry AI delivery, and teaches graduate deep learning at UCLA, as chief technology officer. The people who decide what an agent may do inside a chart have spent their careers accountable for what goes in one.

Everyone else asks you to trust the automation.
We hand you the tape.

Every Meros agent run produces three artifacts before anyone asks for them: a video recording of exactly what the agent saw and did, a provenance record pinning the precise workflow version that ran, and an approval gate where a human signs off before anything lands in a chart. It is the difference between an AI vendor and an auditable operator.

Every run is recorded

Not a log line — a screen capture of the agent doing the work, captured on every fire and retained with the run. When a result looks wrong, you don't open a support ticket and wait: you review the recording of the run that produced it.

Pinned, not improvised

Each workflow is learned once, verified against captured evidence, then pinned by content hash and promoted deliberately. Production replays the exact approved version — pin + hash — so a run is reproducible instead of a fresh guess every night.

A clinician still decides

Agents gather, draft and stage. They do not sign, diagnose or prescribe. Anything clinical routes to the responsible provider for review — and the automation records that it waited.

How it ships

Mapped, verified, pinned, then trusted with production.

Nothing reaches a live chart because it worked once on a demo account. Every agent walks the same four gates, and the gate it passed is part of its record.

Step 01

Map

We sit with the person who does the task today and capture the real sequence — including the exceptions they never wrote down.

Step 02

Learn

The agent performs the workflow under observation, recording every step and screen so the captured run becomes the specification.

Step 03

Verify

The recorded run is checked against the evidence, then pinned by content hash. An approved version is promoted explicitly — never silently.

Step 04

Run

Production replays the pinned version on schedule. Drift is detected, a repair is proposed for approval, and the fleet's health reports itself daily.

HIPAA-aligned by construction

Protected health information stays inside a business-associate boundary. Services outside it never receive patient data — that boundary is enforced in the build, not promised in a policy.

Every run is an audit artifact

Run identifier, workflow version, timestamps, what was read, what was written and who approved it — retained on a defined schedule and produceable on request.

Failure is loud, not silent

A stalled agent raises an alarm and a digest the same day. The failure mode we design against isn't a wrong answer — it's an automation that quietly stopped and nobody noticed.

You keep the keys

It runs in your systems, on your credentials, under your control. Turn an agent off and the workflow reverts to your staff — there is no lock-in and no data hostage.

Leadership

Clinicians and engineers, in the same room.

Two practising physicians, an operator and a computational scientist. Full biographies, credentials and training are on the leadership page.

Founded: 2024
Focus: Specialty medical practices
In production: Pain management practices, Nevada
Engagement: We earn when you earn or save

Meros operates as a small senior team. The person who maps your workflow is the person who builds the agent and the person accountable when it fails. You work directly with the engineers and clinicians responsible for the system.

Meet the leadership team