Genzeon Platform

Genzeon Platforms 🧠

Building the Healthcare Brain — governed clinical intelligence for payers, providers, and government healthcare.

Hi, and welcome. We're the AI platforms team at Genzeon, a healthcare technology company that's spent two decades close to the messy reality of clinical data. This org is where we ship the open pieces of our work: small, fast, permissively-licensed clinical NLP models and the open specs that sit underneath our platforms.

We build for a domain where being right, auditable, and defensible matters as much as being clever — prior authorization, utilization management, revenue cycle, member engagement, and privacy operations. A lot of that is proprietary. The parts that are most useful to the wider community, we open.

🔬 What we've open-sourced

Healthcare Brain — clinical NER family

The Healthcare Brain NER family is our set of production-grade clinical Named Entity Recognition models. They're built on trusted biomedical bases (Bio_ClinicalBERT, PubMedBERT), fine-tuned on clinical corpora, and released under Apache-2.0 so you can actually use them in real systems. Each ships with a live demo Space you can try in the browser.

ModelExtractsBaseLabelsLicenseDemoHealthcare Brain Medication NERDrug names, dosages, routes, frequencies, administration details (12 categories)Bio_ClinicalBERTBIO · 25Apache-2.0Try it →Healthcare Brain Laboratory NERLab tests, values, units + rule-based numeric extractionBio_ClinicalBERTBIO · 21Apache-2.0Try it →Healthcare Brain Diagnosis / ICD NERDiagnoses & conditions, with ICD-10 / SNOMED linking via SapBERT (9 categories)PubMedBERTBIO · 19Apache-2.0Try it →Healthcare Brain Clinical Findings NERSymptoms, findings and clinical mentions from free-text notesBio_ClinicalBERT · spaCy—Apache-2.0Try it →Healthcare Brain Vitals NERBlood pressure, heart rate, temperature, SpO₂ and other vital signsBio_ClinicalBERT—Apache-2.0Try it →Healthcare Brain NER (base)PHI / PII entities for de-identification before downstream useBio_ClinicalBERT—Apache-2.0Try it →

All models are ~110M parameters, use a 512-token window, and run on standard HuggingFace Transformers — small enough to deploy on-prem or at the edge, which matters a lot for PHI. The Diagnosis / ICD model runs a two-stage pipeline: NER extraction, then entity linking for automated code mapping.

A note on models we build on: for healthcare and government contexts we deliberately restrict our open-weight bases to US/EU-origin models. It's a sovereignty and trust choice, and it's baked into how we pick every base you see here.

AKPS — Aether Knowledge Pack Specification

AKPS is our open format for expressing healthcare coverage policy as portable, machine-readable knowledge packs — so coverage logic isn't trapped in any one vendor's system. Released April 2026, dual-licensed CC BY 4.0 (spec) + Apache-2.0 (tooling). If you work on coverage policy, prior auth, or clinical rules, we'd love your eyes on it.

🏥 The bigger picture: the Healthcare Brain

Our open models are the visible tip of a larger system. Genzeon Platforms runs three production platforms on one patented architecture — what we call the Healthcare Brain:

HIP One — the reasoning lobe. Clinical decisions, medical necessity, prior authorization, utilization management, and revenue-cycle intelligence. PES One — the engagement lobe. Conversational AI for member and patient engagement, care navigation, and quality (Stars / HEDIS). CPS One — the governance lobe. Privacy program operations, breach response, and AI governance.

All three run on Aether One™, our patent-protected substrate (USPTO filings in progress). The NER models above are the kind of grounded, auditable NLP that feeds that system.

✅ Where we are

We don't like claiming things we haven't seen run. A few that we have:

Live in production in the CMS WISeR Model (2026–2031) via Novitas in New Jersey — real prior-auth volume at 100% of the CMS 3-day turnaround, with zero automated denials (every denial is signed by a named, licensed reviewer). HITRUST r2, SOC 2, and HIPAA aligned. Partnered with Microsoft (10-year strategic partnership, 5 solution designations), UiPath, and Databricks. Backed by a portfolio of USPTO patent filings on the Aether One™ architecture.

🤝 How to engage

Try the demos linked above — they run in the browser, no setup. Use the models in your own pipelines (Apache-2.0 — go for it). Issues and feedback on any repo are welcome. Look at AKPS if you work anywhere near coverage policy or clinical rules. Building in healthcare AI and want to compare notes? Reach out through genzeon.one.

🔗 Find us

🌐 Website — genzeon.one 📰 Newsletter — Building the Healthcare Brain on LinkedIn 🤗 Hugging Face — @genzeonplatform

Genzeon Platforms — A Genzeon Company. Models and specs here are released under their stated open licenses; our commercial platforms are separate. We think being right and being auditable beat being flashy.