TrellovaTrellovaTalk to us
How it works

One session. Five stages. Closed loop.

Watch a single behavioral-health visit travel from raw audio through to a recovered denial dollar. Every step is verifiable by a human, every claim cites source material, and nothing leaves your tenant's encryption boundary without explicit authorization.

The Trellova workflow loop: capture, draft, submit, resolve, recover.CaptureAmbient audioDraftBH-native noteSubmitX12 278 PAResolve835 + denial detectRecoverLLM appeal letterSessionto dollars

Stage 1
Capture

The clinician starts the session. We listen — encrypted, end-to-end.

Ambient audio streams through LiveKit over secure WebSockets, transcribed by Deepgram Nova-3 with diarization for clinician vs. patient voices.

The browser extension and patient app both work. Telehealth or in-person, EHR-aware or not — the capture surface is identical.

No microphone access on a clinician's laptop without explicit per-session consent; the indicator is always visible. Patient app captures via the same path.

LiveKit WebRTCDeepgram Nova-3 ASRTLS 1.3HMAC-signed bearer auth
Stage 2
Draft

60 seconds later, a source-linked clinical note.

Claude Sonnet 4.6 produces the structured note in the BH templates clinicians actually use — SOAP, DAP, BIRP — with MSE typing, suicide-risk classification, and MBC trajectory baked in.

Every sentence in the note carries a citation back to the exact second in the transcript that produced it. Clinicians verify; they don't rewrite.

Coding — CPT (90791, 90834, 90837, 96127, etc.) and ICD-10 — emerges from the same draft. A junior coder pass on Haiku 4.5 disambiguates the borderline ones.

Claude Sonnet 4.6Claude Haiku 4.5 (coder)Citation-grounded chain-of-thoughtBH-template library
Stage 3
Submit

Before sign — payer pre-flight + auto prior-auth.

The same draft hits the pa-engine, which compares it to the specific payer's published medical-necessity criteria (Cigna's BHQRC, UHC's CAG-A06, Aetna's #0264). Missing interpreter docs? Functional impairment not stated? We tell the clinician before they sign — not after the 30-day denial.

Approved drafts submit through Availity as X12 278 with the supporting documentation pre-attached. Approval-probability shows live on the clinician's screen.

Starting 2027, direct payer FHIR PA APIs replace clearinghouse hops where CMS-0057-F mandates them.

Availity X12 278Payer rules libraryFHIR PA APIs (2027)Approval-probability model
Stage 4
Resolve

Days later, an 835 lands. We classify automatically.

The same Availity channel ingests 835 ERA remittances. The denial classifier sorts every CARC/RARC into one of our 14 canonical denial types — medical-necessity, coding error, eligibility, timely-filing, authorization missing, and so on.

Recoverable denials route to the recovery queue. Hard denials (true non-covered services) get a clinician note explaining why we won't appeal.

X12 835 ERA ingestCARC/RARC classifier14-type denial taxonomyAuto-reconciliation
Stage 5
Recover

Claude Opus 4.7 drafts the appeal. Clinician signs in one click.

For each recoverable denial, we cite the payer's own published criteria, the transcript evidence (with timestamps), the matched documentation in the note, and any prior-payer-acceptance precedent from your tenant.

Appeals go back through Availity. Tracking is automatic; the clinician sees status without leaving the EHR.

Every recovery sharpens the appeal-pattern library — the model learns which arguments win against which payers for which codes, across the whole customer network (PHI-stripped).

Claude Opus 4.7Appeal-pattern libraryPer-payer precedent matchingNetwork-effect learning

What the clinician actually does

The product is built around the principle that the clinician's attention is the most expensive thing in the room. We minimize their workflow surface; we don't add to it.

Pre · Open the EHR like normal

Trellova sits in a sidebar via the browser extension, or runs as a separate tab on EHRs without an extension surface. Patient eligibility prefetch happens at scheduling time — we already know the payer's PA requirements before the visit starts.

Mid · Have the conversation

Mic-on indicator stays visible. Live co-clinician suggestions appear in the sidebar — "consider documenting functional impairment for 90837" — without interrupting the session.

Post · Verify, sign, move on

~3-minute review: scan the note, click citations to verify anything ambiguous, sign. PA submits automatically; denial appeals draft automatically; the clinician sees a summary of outcomes in their daily digest. That's it.


The time math

Behavioral-health clinicians lose ~13 hours per week to documentation and prior-auth paperwork. Trellova returns ~10 of those hours.

~18 hrs/wk
Today

Per clinician on documentation + prior auth — 30% of working hours patients never see.

~5 hrs/wk
With Trellova

Verification, sign-off, and the occasional appeal review. The rest happens in the background.

+65 hrs/wk
Practice-wide (5 clinicians)

≈ 1.6 additional clinician-FTE of session capacity without hiring a new clinician.


What we don't do

Equally important. We sit alongside the products you already own — never displacing them.

  • Be the EHR. We don't do scheduling, claims billing, or patient portal.
  • Be the clearinghouse. We use Availity; we don't replace it.
  • Train on your data. Per-tenant data stays in your encryption boundary. Anthropic + Deepgram + LiveKit hold BAAs but model training is off by contract.
  • Talk to payers in your name. Every PA submission, every appeal — clinician-signed and clinician-reviewed before it goes out.
  • Replace the clinician's clinical judgment. We surface evidence; the clinician decides.
  • Charge surprise fees. Subscription + 4% of recovered denials. That's the whole price list.

Want to see this on real audio?

We're booking 30-minute demo calls with practice owners and clinical leads. We'll run a sample session live and walk you through PA + appeal for a denied claim.