Penguin Ai

Penguin Ai's workers use language models to extract chart evidence, apply payer policy and prepare coding, authorization and claims work. The company also sells its data and context layers separately. Those layers organize records and policies, while the worker layer depends on learned reasoning to complete its core tasks.

Review checkpoints can be configured by case, sample, risk or dollar threshold. Uncertain results route to reviewers, and clinical determinations require a person. Revenue cycle specialists can dismiss, reassign or escalate findings, retain final approval and control submission, while operators can change the thresholds governing the workflow.

Gwen's language service runs through AWS Bedrock and identifies Sonnet as its default model family. Specialized pipelines add evidence extraction and validation, while skills can be pinned to a version. The exact production model release and training data for each worker are not specified.

AWS Bedrock and the Sonnet model family are identified for Gwen's language service. Penguin Ai also describes its own models and customer account execution through Snowflake or AWS. The complete processor inventory and model specific retention terms are not available across all these deployment paths.

The prior authorization model card reports an F1 score of 0.954 for the Gwen pipeline with GPT 5.5, compared with 0.918 for the model alone. The benchmark does not state its sample count, reference labels or evaluation date. Operational percentages likewise lack a named study population and measurement protocol.

Customer data is excluded from shared model training, while learning from an organization's operations stays within that organization. Gwen limits ordinary testing to synthetic data. Retention periods and deidentification rules are unspecified. Evidence grounding and confidence routing support review, but a complete clinical safety incident protocol is not described.

Penguin Ai states HIPAA compliance and signs BAAs for customer relationships. Gwen requires Enterprise deployment and security review before real patient data can be used. The privacy notice centers on contact, account and website information; the service PHI agreement and its subcontractor obligations are not available to read before contracting.

SOC 2 auditing is stated for Penguin Ai, with further assurance material available to prospective customers under NDA. Encryption, permissions and audit logs are also described. The public pages do not identify the SOC report type, examination period or scope, and the restricted report remains unavailable for direct inspection.

The workers are positioned for healthcare administration, including coding, claims and preparation of authorization decisions. Clinical reasoning is routed to human reviewers, who retain the decision. The product does not present itself as an autonomous diagnostic or treatment device, and no FDA clearance is claimed for these workflows.

Skills are evaluated against reference datasets, with accuracy, confidence calibration, cost and latency tracked. Customers can pin skill versions, and policy changes trigger validation of affected workflows. This provides a defined governance process; performance comparisons across demographic or clinical subgroups are not specified for the deployed workers.

Penguin Ai advertises a 90 day return on investment guarantee for engagements and says systemic errors are corrected, communicated and documented. Glassbox connects findings to their supporting evidence. The guarantee's eligibility rules, calculation and remedy are not specified, and the general service terms remain an unfinished draft.

Epic FHIR and Cerner are named connections, alongside HL7, FHIR R4, X12 and NCPDP support. Browser automation reaches portals without modern APIs, and workers produce structured output for downstream systems. Named production evidence of a major EHR connection that both reads and writes is not supplied.

Native applications can run within a customer's Snowflake or AWS account, and dedicated private VPC deployment is supported. The Snowflake offering keeps records, recommendations and audit activity in that account. A complete regional availability list and processing geography for the hosted Gwen service are not specified.

Gwen offers a free trial allocation and a Pro plan at $450 per month with 250 monthly credits. Usage depends on tokens, with additional credit purchases available. Real patient data requires an Enterprise agreement and security review; its production price and the wider operating system's minimum commitment are not specified.

The platform targets health plans, health systems, physician groups and healthcare service organizations, with dedicated home health coding and OASIS workflows. Its coding catalog also covers orthopedic and evaluation and management tasks. Settings and workflow names are clear, but validation across those care settings and specialties is not established.