RADV
audit readiness, year-round
Regulatory-grade RADV
Mock audits with RADV-style sampling and extrapolation exposure modeling. See which HCCs survive before CMS does — evidence packets in one click.
View RADVRegulatory-grade risk adjustment.
Run inside your environment.
From chase list to submission-ready files: Martlet AI closes 95% of cases automatically at 99% precision, links every HCC to page-level evidence, and packages CMS-ready RADV audit packets. PHI never leaves your network.
- 95%
of cases closed automatically, end-to-end
- 99%
precision on every code the model automates
- 95%
reduction in chart review time
- 100,000+
lives served by our platform
- 10
health systems and payers in production
- Zero
PHI leaves your network — on-prem, VPC, or air-gapped
Martlet AI covers all three risk-adjustment workflows.
RADV, retrospective, and prospective — each covers a different part of the cycle, and you can run one or all three.
Retrospective
high volume
95% of cases closed automatically
Prioritized chase lists, undercoded HCC capture, MEAT validation, and submission-ready deltas. Reviewers work exceptions, not stacks of PDFs.
View RetrospectiveProspective
point of care
Suggestions clinicians act on
Suspected conditions from structured and unstructured data, delivered pre-visit and in-visit as concise, evidence-linked suggestions in your EHR.
View Prospective
RADV — verify every code the way CMS verifies it.
A detected encounter. A credentialed provider. A signed note. MEAT-supported documentation. Martlet AI runs the checks CMS applies at audit — automatically, on every chart — and links page-level evidence to every HCC. Codes that pass close on their own. Codes that don't get caught before CMS catches them.
progress_note_2025-03-14.pdf · page 4
Reviewed labs and medication adherence. A1c 8.1%. Type 2 diabetes with diabetic polyneuropathy — continue metformin, titrate gabapentin. Follow-up in 3 months. Discussed diet and foot care; monofilament exam performed.
ICD-10 E11.42 → HCC 37
Diabetes with chronic complications
- EncounterOffice visit · detected
- Date of service03/14/2025
- ProviderJ. Rivera, MD · credential verified
- SignaturePresent · e-signed
- MEATEvaluate · Treat
Closed automaticallyconfidence 0.97
All patient data shown is synthetically generated for illustration.
- Encounter detection
- Signature verification
- Provider credential check
- Page-level evidence links
Retrospective — from your data to your submission file.
Charts, claims and prior coding go in. Verified codes and submission-ready files come out, with no hand-offs to vendors or spreadsheets in between.
Ingest
Claims, EHR notes, scanned PDFs — normalized and de-duplicated.
Prioritize
Charts ranked by expected RAF impact and documentation strength.
Validate
Every code checked against the record, evidence linked to its page.
Review
Exceptions routed to your coders, with the evidence already attached.
Submit
Adds and deletes exported in the format your submission pipeline expects.
Everything a risk-adjustment program runs on.
Every capability behind RADV, retrospective, and prospective — from ingestion to submission.
- 01
Chase list prioritization
Chases ranked by expected RAF impact and documentation strength. Work the charts that matter first, and track every chase to closure.
- 02
CMS requirement verification
Encounter detection, provider credential checks, and signature verification on every chart — the checks CMS coders run, run first.
- 03
MEAT validation
Monitor, Evaluate, Assess, Treat — validated at the sentence level, with v24/v28 dual mapping and payment-year discipline.
- 04
Mock RADV sampling
RADV-style sampling and extrapolation modeling against your own contracts. See your exposure before CMS measures it.
- 05
One-click evidence packets
Chart sentence, encounter ID, date of service, provider, credentials, signature — assembled into a CMS-ready packet per HCC.
- 06
Submission-ready outputs
Adds and deletes exported in the format your submission pipeline expects. Validated codes flow to submission; unsupported codes flow out.
- 07
Reviewer UI + exception queues
The 5% that needs judgment lands in queues with linked evidence. QA sampling and dual review built in.
- 08
Complete audit trail
Who coded what, when, and based on which evidence — every action recorded, ready to hand to an auditor.
- 09
RAF analytics
RAF lift, capture rate, coder throughput, and cost per chart — measured continuously, reportable to the CFO.
- 10
EHR + claims ingestion
FHIR R4, HL7, CCDs, claims extracts, and scanned PDFs with OCR. Millions of charts, normalized and de-duplicated.
- 11
Suspecting engine
Suspected conditions surfaced from structured and unstructured data — labs, meds, notes — prioritized for pre-visit review.
- 12
Your deployment, your terms
On-premises, private cloud (AWS, Azure, GCP), or air-gapped. Your existing security controls stay in effect, because nothing leaves them.
Built for in-house, end-to-end, audit-grade work.
Regulatory-grade accuracy
Healthcare-specific AI, not a wrapper
Martlet AI runs on John Snow Labs' production medical language models — ranked #1 on 12 of 13 medical benchmarks against frontier general-purpose LLMs. Every HCC links to MEAT evidence at the chart-sentence level, so every output is reproducible and defensible under RADV.
See how accuracy holds up under auditRun in-house
Runs inside your environment, under your controls.
Deploy on-premises, in your private cloud, or air-gapped. No external API in the data path, PHI that never leaves your network, and no services contract paid on success commission — you keep the coding, the knowledge, and the controls.
How deployment worksAutomation, not assistance
95% closed automatically. Reviewers see the exceptions.
Martlet AI runs at 99% precision on the codes it automates — high enough to close 95% of cases end-to-end and cut review time by 95%. Coders see only the cases that need judgment.
See the workflow
In production at health systems, payers, and risk-bearing providers.
Featured case studies
Arkos Health runs Martlet AI across the full risk cycle: retrospective recapture, prospective capture at point of care, and mock RADV testing with automated evidence packets.
Incremental Revenue
Meaningful incremental revenue from validated HCC recapture + in-year capture
HCC Coding Accuracy
Higher coding throughput by reducing manual chart hunting and the rework
Improved Audits
Improved audit readiness through mock RADV testing & evidence packet automation
Partners
Built for the review that happens three years later.
RADV auditors read every output long after submission. Martlet AI is engineered so each one holds up.
HIPAA-aligned, in-environment deployment
On-premises, private cloud, or air-gapped. PHI never leaves your network, and your existing security controls stay in effect.
Governed AI, not a black box
Every model versioned, every release tested before it touches production charts, every decision explainable — AI your compliance team can sign off on.
An answer for every auditor
Every code, every change, every reviewer action — recorded. When CMS asks why a diagnosis was submitted, the answer is one click away.
Peer-reviewed accuracy
30+ published papers behind the underlying medical language models. Benchmarks you can read, not adjectives.
How the engagement works.
An annual license
You pay a license fee for the year. Not per chart, not per code we validate, and not a percentage of what you capture.
Priced to your size
The fee is based on how many contracts and lives you run, and stays fixed for the year. Add contracts, volume or payment years without renegotiating.
Configured to your process
We set the review levels, thresholds and exports to match how your team works, and connect to the systems they already use.
Questions buyers ask first.
What is regulatory-grade HCC coding?
Regulatory-grade HCC coding means every code is verified against the requirements CMS applies at audit: a valid, detected encounter; a credentialed provider; a present signature; and MEAT-supported documentation — with page-level evidence linked to each HCC. Martlet AI runs these checks automatically on every chart, so submissions are defensible under RADV three years later.
How does Martlet AI close 95% of retrospective cases automatically?
Martlet AI's healthcare-specific medical language models validate each HCC against MEAT criteria and CMS requirements with a calibrated confidence score, operating at 99% precision on the codes it automates. High-confidence cases — 95% of the total — close end-to-end; the rest route to reviewer queues with linked evidence. The result is a 95% reduction in chart review time.
Does PHI ever leave our network?
No. Martlet AI deploys on-premises, in your private cloud (AWS, Azure, or GCP), or fully air-gapped. There are no external AI API calls in the data path, and PHI never leaves your network — your existing security controls, SIEM, and IAM stay in effect.
What does Martlet AI hand off to our submission pipeline?
Adds and deletes, exported in the format your submission pipeline already expects, each carrying the date of service, the provider and the linked evidence behind the code. Unsupported codes identified during validation flow out as deletes, closing the loop on RADV exposure.
Does Martlet AI support the v28 transition?
Yes. Martlet AI maps every diagnosis under both CMS-HCC v24 and v28 with payment-year discipline: codes are validated against the model that applies to the payment year in question, and outdated codes are blocked at submission.
How do we prepare for a RADV audit?
Run mock RADV audits year-round: Martlet AI samples your contracts RADV-style, validates each sampled HCC against MEAT and CMS requirements, models your exposure, and assembles CMS-ready evidence packets — chart sentence, encounter, provider, credentials, and signature — in one click. When CMS selects your contract, the packet is already built.
Bring an anonymized chart. We'll show you what audit-grade looks like.
Bring an anonymized chart. We'll run Martlet AI on your data, walk the retrospective and RADV workflows end-to-end, and answer the security and deployment questions your team will ask.
- 95%
closed automatically
- 99%
precision on automated codes
- 100,000+
lives on the platform