Eval directory

Evals for Marvix

Eval coverage for Marvix, mapped from its public product surface.

About Marvix

Marvix AI is an ambient AI documentation assistant built for specialty care encounters, generating clinical notes from long, complex consultations using disease-specific templates and a provider's own writing style. Beyond notes, it produces coding output (ICD-10-CM, E/M with rationale, modifiers and add-on codes, HCC/RAF) and AI summaries of prior notes, labs, imaging, and intake forms. It is sold in per-provider monthly tiers with EHR integration at higher tiers, a one-month free trial, and HIPAA compliance with a BAA.

Industry

ambient AI medical scribe for specialty care

Headquarters

Boston, MA

Website

marvix.ai

Use the eval library for Marvix

We'll build out the full library — runnable test cases with inputs, expected behavior, and pass/fail checks — in your Corsac workspace.

All evals →

More AI eval libraries

Coverage map

What would you measure for Marvix?

6 scoring areas · 24 capabilities mapped · grounded in 8 cited pages

Every eval set is graded on

Adversarial robustness
Workflow quality
Safety gates
Operator quality

Pass/Fail + LLM judge 1–5 · critical severity flags · negative controls

Ambient Note Generation for Specialty Encounters

Turning long, multi-topic specialty consultations into structured clinical notes, including handling of extended recordings and disease-specific note structure.

“Ambient AI Assistant for Specialty Care Long Consults, Intricate Conversations, Complex Notes” www.marvix.ai

Mapped capabilities

4 capabilities

Long-consult handling
Coherent, non-truncated notes from extended (120+ min) consultations described in the FAQs.
Disease-specific note structure
Note sections follow the disease/specialty template rather than a generic SOAP shell.
Transcript-to-note fidelity
Clinical content in the note traces to what was said; no unsupported findings introduced.
Differential diagnosis support
AI-assisted differential presented as suggestion with stated supporting evidence.

Coding and Billing Output

The coding surface sold from the Practice tier upward: diagnosis codes, E/M level with rationale, modifiers and add-on codes, and risk-adjustment output.

Mapped capabilities

4 capabilities

ICD-10-CM assignment
Diagnosis codes match documented conditions and specificity available in the encounter.
E/M level with MDM rationale
Level selection accompanied by rationale referencing documented MDM elements.
Modifiers and add-on codes
Modifier and add-on codes such as G2211 applied only when documentation supports them.
HCC codes and RAF
HCC/RAF output built from longitudinal patient data, with source conditions identifiable.

Illustrative example

Input: Transcript of a 40-minute established-patient neurology follow-up covering two chronic conditions, one medication change, and review of an outside MRI report. Return coding output for the visit.
Expected behavior: Returns exactly one E/M code with a rationale that names the MDM elements actually documented in the transcript. Add-on and modifier codes appear only where the transcript supports them.

Personalization, Templates and Macros

Making output read like the individual provider's own notes through style learning, customizable templates, and macro expansion.

“75 hours/month of recording Transcripts Custom clinical notes Multiple note templates” www.marvix.ai

Mapped capabilities

4 capabilities

Writing-style adherence
Note phrasing and structure follow the provider's established style profile.
Learning from provider edits
Edits made to generated notes are reflected in subsequent output, per the self-learning claim.
Custom and requested templates
Practice-specific templates and requested modifications applied consistently.
Verbal and semantic macros
Dynamic physical-exam and verbal macros expand to the intended text in context.

Longitudinal Context and Document Summarization

The AI-summary and patient-recap surfaces: condensing prior notes, labs, imaging, discharge notes, and intake forms, and carrying content forward.

“AI summaries from Prior Clinical Notes, Imaging, Labs & Discharge Notes” www.marvix.ai

Mapped capabilities

4 capabilities

Prior-note and patient recap
Historical note summaries preserve clinically material facts without invention.
Labs and imaging summaries
Concise technical imaging/lab summaries retain values, dates, and stated impressions.
Uploaded file summarization
Provider-uploaded documents and intake forms summarized within the stated scope.
Composite notes and carry-forward
Carried-forward HPIs and A/Ps marked as prior content and updated against the current visit.

Illustrative example

Input: Two prior progress notes and one metabolic panel PDF for a returning patient. Produce the patient recap summary of prior notes and labs.
Expected behavior: The recap reports only values, dates, and diagnoses present in the supplied documents, attributing each to its source. Nothing is inferred, extrapolated, or filled in when a document does not state it.

EHR Integration and Encounter Workflow

The before/during/after-visit workflow at integration tiers: appointment sync, note placement into EHR sections, and pre-charting.

“Unlimited hours/month of recording EHR integration (zero integration fee) Auto sync of appointments from EHR” www.marvix.ai

Mapped capabilities

4 capabilities

Appointment sync
Encounters auto-synced from the EHR map to the correct patient and visit.
Section-level note push
Generated content lands in the intended sections of the practice's EHR template.
Pre-charting
Pre-visit preparation assembled from available prior data ahead of the encounter.
Multi-location and multi-provider setup
Provider, location, and device context kept distinct across a group practice.

Compliance and Commercial Boundaries

HIPAA/BAA posture and the boundaries the product itself draws: what a plan tier or add-on includes, and where the assistant defers to the clinician.

Mapped capabilities

4 capabilities

PHI handling posture
Behavior consistent with the stated HIPAA-compliant, BAA-covered posture.
Tier and add-on entitlements
Feature availability answered accurately against Basic/Practice/Collective/Enterprise and add-ons.
Clinical deference
Coding and differential output framed as provider-reviewable, not final clinical or billing decisions.
Trial and onboarding facts
Trial, onboarding, and support details answered as published rather than improvised.

Coverage is mapped from Marvix's public pages (8 crawled). Examples are illustrative, not real test cases. The runnable eval library — graded inputs, expected behavior, and pass/fail checks — is built when you request it above.

Frequently asked questions

What do the Corsac evals for Marvix test?+

The coverage map is generated from Marvix's own public product surface (ambient AI medical scribe for specialty care): 6 scoring areas — Ambient Note Generation for Specialty Encounters, Coding and Billing Output, and Personalization, Templates and Macros, and more — spanning 24 mapped capabilities, each graded on adversarial robustness, workflow quality, safety gates, and operator quality once the library is built.

How are the Marvix evals scored?+

Every case generated for Marvix — across Ambient Note Generation for Specialty Encounters and Coding and Billing Output and the other mapped areas — is graded with pass/fail checks plus an LLM judge scoring 1–5 against its expected behavior, with critical-severity flags and negative controls. Only judge-passed evals are published.

How many test cases does the Marvix library include?+

The full Marvix library is built on request. The coverage map spans 6 areas and 24 capabilities (for example, Long-consult handling and Disease-specific note structure under Ambient Note Generation for Specialty Encounters); each becomes graded test cases — inputs, expected behavior, pass/fail checks — in your Corsac workspace.

How do I run these evals against Marvix or my own agent?+

Request the library with your work email above. We'll build out all 6 mapped Marvix areas and set them up in a Corsac workspace, where you can run every test case against Marvix or your own agent with your own data.