Reveleer

The model is the engine of the core modules, and the platform carries other value. EVE, the Evidence Validation Engine, is described on the company's own page as a shared AI layer embedded in the platform rather than a product sold on its own: it generates prospective risk suspects, scores retrospective findings, extracts and normalizes clinical evidence and prioritizes charts for coders. Those capabilities do not exist without it.

The rest does. The company began in 2009 as a medical record retrieval business, and the retrieval operation, the relationships and workflows that get charts out of thousands of provider offices at volume, remains a substantial asset alongside managed review services, a clinical data repository launched in 2026, care gap and provider directory tools, and member management.

The model proposes and a person confirms, with one part of the structure unpublished. The EVE page says coding and clinical decisions require human validation, that retrospective findings are ranked and explained for certified coders who review before submission, that prospective suspects are delivered into point of care workflows for clinician confirmation, and that agentic components act within defined rules and remain non autonomous. Every suggestion links back to its source page, so a reviewer can trace each abstracted element. That structure fits an output that becomes a regulatory submission where the plan bears audit liability for every code.

What is missing is the threshold: at what confidence score a finding is shown, suppressed or routed for a second review, and what the AI voice agent used in record retrieval may say to provider offices and when it hands a call to a person.

The engine's parts are named, but the version discipline behind them is not shown. The EVE page names what the Evidence Validation Engine is made of: hybrid AI combining generative evidence extraction with deterministic formulas and clinician authored logic; the EVE Confidence Score, which applies evidence based scoring and supervised machine learning to rank retrospective findings; agentic AI pipelines that extract, validate and route work within defined rules; clinical natural language processing including techniques enhanced by large language models; supervised predictive models; and an AI voice agent used only in record retrieval. It says every insight links to source documentation and that updates are controlled under an MLOps framework.

The site does not say which generative and language models are used or what the supervised models were trained on, and it gives no precision and recall for abstraction or suspecting and no version or change log a customer can follow. For output that feeds RADV submissions, that leaves no way to tell when the model that found a condition changed.

Model services are named for part of the chain, and the newest layer's provider is not. An Amazon Web Services case study, quoting the company's own staff, describes the platform running on AWS and using Amazon Textract to extract text and handwriting from scanned records and Amazon Comprehend Medical to extract clinical information, with conditions then highlighted for coders to review. That names the host and two model services in the chain.

The EVE page has since added generative evidence extraction, techniques enhanced by large language models, agentic pipelines and an AI voice agent in retrieval, and names no provider for any of them; the case study is undated, so whether Textract and Comprehend Medical still carry the load is itself a question. No subprocessor list, and no retention or training terms governing data once it reaches a model service, appear on the home, security, EVE or privacy pages; the privacy policy offers a list of processors on request only for website data.

Vendor figures credited to its own deployments, with no method attached and one internal inconsistency. The EVE page reports a 3X noise reduction, described in the same sentence as cutting false positives and irrelevant suspects in half compared with legacy NLP, and a 33 percent improvement in RAF accuracy that it says EVE's confidence score has helped health plans achieve; a case study on clinical coding precision is linked. An Amazon Web Services case study quoting company staff reports 100 percent platform uptime and 90 percent of responses served in under eight seconds, which is operational performance rather than accuracy.

Every figure traces back to the company. No baseline, denominator, period or named plan accompanies the figures where they appear, and "3X" and "in half" cannot both describe the same reduction. The endpoints are coding yield and review efficiency; nothing published measures whether suggested codes survive audit, which is the outcome that matters to a plan.

The surface is large and quantified by the company, and the architecture has recently made it durable. The company has described processing over 1.2 billion pages of clinical data a year for more than 55 health plans. The clinical data repository launched in 2026 turns episodic retrieval, where a chart is pulled for one purpose, into a persistent member level record held once and reused across risk adjustment, quality, audit and care management, and its page says the repository can feed future AI and predictive models with cleaner, more complete clinical inputs. The EVE page says the engine does not synthesize new clinical content.

What is not published: retention periods for repository charts, whether member records from one plan's contract enrich models used for other plans, what de identification is applied before any model development, and what is returned or deleted at contract end. No safety engineering is published either: how a wrongly extracted condition is caught and corrected before it reaches a submission.

Business associate status is stated and the only privacy document covers the website. The privacy policy, last updated 16 June 2026, says Reveleer handles the health data it processes for customers as a business associate under HIPAA, solely under its customer agreements and business associate agreements, and that those contracts rather than the policy govern how that data is collected, used, protected and disclosed; the policy itself applies only to website and marketing data. The security page adds a HIPAA compliance statement. No business associate agreement, subcontractor flow down or breach notification term is published.

The operating model makes it a three party question. The plan is the covered entity that contracts; the charts are obtained from provider organizations, separately covered entities that are not the customer and hand records to a company they did not choose. Since 2026 the clinical data repository keeps those charts for reuse across risk adjustment, quality, audit and care management, so a chart released for one purpose now persists for others.

A recognized certification is named in the company's own material, without the artifact and with two pages describing it differently. The security page says the company has achieved HITRUST CSF certification alongside HIPAA compliance, and the site footer carries a HITRUST CSF Certified seal. The clinical data repository page, launched in 2026, describes HIPAA and HITRUST aligned controls across the platform, which is a weaker statement than certified, and neither page names the assessment type (the one year i1 or the two year risk based r2), the scope of systems covered, the date or the expiry. The security page describes controls against unauthorized access, integrity protection and threat detection in general terms; the repository page adds audit trails for chart access and role based permissions.

No SOC 2 report and no trust portal is linked. For a platform holding a persistent member level chart repository, whether that repository sits inside the certified scope is the specific question.

No device position is stated, and the suspecting function is documentation with its basis shown. EVE's prospective workflow surfaces evidence backed diagnosis suggestions, possibly undocumented chronic conditions, into point of care record workflows for a clinician to confirm, and its retrospective workflow prioritizes findings for coders; the EVE page says every suggestion links to source documentation, that EVE does not synthesize new clinical narratives or diagnoses, and that coding and clinical decisions require human validation. Surfacing conditions with the chart evidence shown for someone to confirm is a documentation function with a reviewable basis, not a diagnostic claim.

The regime that governs the product is payer documentation and audit law rather than device law: CMS risk adjustment rules, RADV audits including the independent validation stage, and NCQA HEDIS abstraction. Building for those regimes is not a regulatory pathway, and no FDA position is stated on the home, security, EVE or privacy pages.

A governance framework with named process, and no results from it. The EVE page says all of the engine's capabilities operate under a structured MLOps and governance framework with continuous monitoring, bias and drift checks, controlled updates and mandatory human validation for coding and clinical decisions, and that agentic components act within defined rules and stay non autonomous. Named monitored checks and a controlled update process are more than values language, and the framework is specific to the product. No bias evaluation, subgroup result or framework document is published, which is what A would need.

Two risks deserve the evaluation. Parsing variability: records come from thousands of provider organizations with very different systems and documentation quality, so a model that reads some formats reliably and others poorly finds conditions unevenly, and the gap follows how well resourced the practice is rather than how sick the member is. And direction: finding an additional condition raises the plan's payment, so the commercial gradient runs toward adding codes, and the company sells both the capture and the audit defense.

Mechanisms exist that let a reviewer challenge an output, and nothing stands behind it. The EVE page says every suggestion links to the source documentation it came from, that each suggestion is explained so coders can check it, and that coding and clinical decisions require human validation; the confidence score ranks retrospective findings by how well the evidence supports them. That is source traceability and review before commit, and the evidence trail is also what a plan produces to an auditor.

No warranty, indemnity, remediation commitment or published error rate appears on the home, security, EVE or privacy pages. The consequence falls on the plan: output feeds risk adjustment submissions and RADV audits, where an unsupported code is recovered by the federal program and a pattern of them becomes the plan's compliance problem, while a missed supported condition leaves the member's record incomplete.

Integration is described in categories and no system is named. The EVE page says prospective insights are delivered directly into point of care record workflows, and the company's clinical data repository page says it integrates with record systems, health information exchanges, retrieval vendors and internal data warehouses, with charts moved by bulk upload, SFTP or API. The distinctive capability is breadth of retrieval across the provider landscape rather than depth with any one system.

No record system, exchange, interoperability standard or certification is named on the home, security, EVE or privacy pages, so a plan cannot tell which point of care systems the suggestions actually reach or by what standard.

A single hosted platform on a named cloud, with no location committed. An Amazon Web Services case study quoting company staff describes the platform running on AWS; the privacy policy names AWS, processing in the United States, only for the website. No region, tenancy model, backup posture or residency commitment for the product appears on the home, security, EVE or privacy pages.

The gap matters because of what the platform now holds. A persistent member level repository serving competing health plans in the same markets raises the tenancy question directly: member populations overlap, since a person changes plans while their medical history does not, and nothing published says whether records are partitioned strictly by contracting plan or how a chart gathered under one plan's contract is treated when the member moves. Retrieval has its own footprint, from provider portals and credentials to on site collection.

Pricing is unpublished, though the way the product is sold is spelled out. The company offers platform software and end to end managed services for record retrieval and review, two economically different models: software typically prices per member, per plan or per module, and a managed service per chart retrieved, per record reviewed or per full time equivalent replaced. A plan comparing vendors may be comparing different units without either side saying so.

One possibility is specific to risk adjustment, where a vendor is sometimes paid a share of the incremental payment its work produces. If any part of an arrangement is contingent on conditions found, the fee runs the same direction as the coding, and the party absorbing an over captured condition is the federal program rather than the plan. The company does not say whether contingent terms are offered, and it publishes no fee at risk against a measure.

The work is records and data handling for health plans and risk bearing providers, covering medical record retrieval, HEDIS abstraction and provider directory validation.

Broad across the payer quality and risk surface, and no longer only retrospective. The home page lists member management, prospective risk adjustment, quality improvement with care gap management, HEDIS audit and provider directory validation, and retrospective risk adjustment with RADV audit support, sold to payers and to providers and ACOs, across Medicare Advantage, ACA Marketplace and Medicaid programs. The prospective work reaches the point of care, which extends the product from look back chart review toward suspecting during the care cycle.

Buyers are health plans and risk bearing provider organizations rather than clinicians, and the setting is the plan's operations and the provider's record rather than any clinical facility. The prospective side is newer than the retrieval and retrospective business, and nothing published shows how widely it runs.