Custom AI/ML solutions for healthcare & life sciences industry

Yes — we build to the FDA's Good Machine Learning Practice principles from day one. Models intended for SaMD submission ship with a predetermined change control plan written to the FDA's August 2025 final PCCP guidance for AI-enabled device software functions, full validation packs against holdout cohorts, and the documentation needed for 510(k) or De Novo pathways.

Both — depending on what you want to own. Most clients integrate directly with their Epic / Oracle Health EHR (formerly Cerner Millennium) / MEDITECH via FHIR + HL7 so models read live data and write decisions back to the EHR. For research-grade work (drug discovery, trial feasibility), we typically run alongside the EHR on a separate data lake. Either way, you own the integration code and the model artifacts.

Yes for the underlying ML platform (feature store, model registry, monitoring, governance) — those are reusable across all three. The domain models differ: hospital ops models target clinical and operational outcomes, payer models target risk and claims, pharma models target target-discovery and trial feasibility. Most integrated delivery networks running their own plan run one shared platform with three model families on top.

Yes, and the deadlines are already live. Since January 1, 2026, impacted payers (Medicare Advantage organizations, Medicaid and CHIP fee-for-service and managed care plans, and QHP issuers on the federally facilitated exchanges) must decide expedited requests within 72 hours and standard requests within seven calendar days, give a specific reason for every denied non-drug request whatever the channel, and publicly post prior authorization metrics. The first reports, covering calendar year 2025, were due March 31, 2026, and refresh every March 31 after that. The FHIR Prior Authorization API follows on January 1, 2027. We build drafting and denial-prevention models on the same FHIR resources, so the letters your staff sends and the payload your API returns come from one pipeline.

Abridge and Microsoft Dragon Copilot (the March 2025 merger of Dragon Medical One and DAX Copilot) are packaged ambient scribes, sold per clinician with the vendor hosting the model and deep Epic integration; Abridge took Best in KLAS for Ambient Speech in 2025 and 2026. A custom build covers the parts that sit outside a packaged product: specialty templates matched to your service lines, note formats your coders and payers accept, write-back into a non-Epic EHR, and inference inside your own tenant so audio and transcripts stay on your network. You own the prompts, the evaluation set, and the integration code.