Showcase · A governed agentic platform, built
See what an agentic
platform can be.
“Agentic automation” stays abstract until you watch one work — so we built one. Meridian is a governed prior-authorization co-pilot: an agent does the judgment-heavy clinical work, a person makes the final call, and every action it takes is checked against exactly what it’s allowed to do.
Why we built it
Most business leaders can’t picture what “an agentic platform” actually is — or why the word governed is the part that matters. A demo that just chats doesn’t help. So we built a real one, in a workflow anyone in healthcare already understands: prior authorization — the review that decides whether a treatment gets approved.
Meridian shows both halves of what we do at once: an agent doing genuine, judgment-heavy work — and the guardrails that make it safe to put live. It’s the fastest way to see the art of the possible and the discipline underneath it, in one place.
Reference demonstration
Meridian runs on synthetic data and was built to show what we build — and it’s proven end-to-end on real infrastructure, not a mockup. It is a demonstration, not a shipped client system. The same governance you see here is the standard we build into every engagement.
The workflow
The agent does the heavy lifting. A person makes the call.
One prior-auth review, start to finish — the AI handling the reading, reconciling, and drafting, with a human at the decision that carries consequence.
- AI 01
Read the member’s chart
The agent takes in a synthetic patient record — diagnoses, history, the request under review — the messy, unstructured input a human reviewer starts from.
- AI 02
Retrieve the governing policy
It pulls the specific medical-policy criteria that decide this case, so the recommendation is grounded in the rules — not in the model’s general knowledge.
- AI 03
Draft an approve-or-deny recommendation
It writes a recommendation with the evidence cited line by line and a calibrated confidence — the reasoning a reviewer can check, not a black-box verdict.
- Human 04
A person makes the final call
Nothing is final until a human reviewer signs it. The agent does the heavy lifting; the accountable decision stays with a person.
What “governed” looks like
Three moments. One that does the work — two that stop it cold.
The whole point of a governed platform is that the interesting moments aren’t just when the AI succeeds — they’re when it’s prevented from doing something it shouldn’t. Meridian demonstrates all three, live.
It does the work, then hands you the decision
On a clean case, the agent runs the whole review — chart to cited recommendation — and parks it for a human to approve. Minutes of expert judgment, drafted and evidenced, waiting for a signature.
Ask for a member it wasn’t assigned → denied
Request a patient this task isn’t scoped to, and it’s refused at the boundary — before any model even sees the data. The agent can only ever touch exactly what its task named. Least privilege, enforced, not promised.
Try to send raw patient data to the cloud → blocked
When a step would route raw PHI to a cloud model, it’s stopped and recorded — the sensitive data never leaves the secure boundary. Data residency by construction, so a mistake becomes a blocked, logged event instead of a breach.
How it’s built
The same governance we build into everything.
Meridian isn’t a one-off. It’s built on the same foundation we bring to a client engagement — which is the point: the guardrails you see here are our standard, not a special case.
Built on Nebula
Our own build playbook — the sixteen proven blocks we use to stand up production AI. Meridian is what they compose into, governance included.
Zero ambient authority
The agent holds no credentials and runs inside a hardened isolation boundary. It gets tools, not tokens — so even a fully hijacked agent has nothing to misuse.
A complete, readable trail
Every decision and action — who acted, on whose behalf, on what, and why — is recorded as one export a compliance reviewer can actually follow.
Want one of these for your business?
Meridian is a demonstration — the real version is the one we build around your workflow, your data, and your compliance bar. Tell us the process you’d want an agent to run, and we’ll show you how we’d build it, governed from day one. No decks.