Creator of OpenMed

Maziyar Panahi

Building open, on-device medical AI — OpenMed · Arcee AI · CNRS

I created OpenMed: open-source medical AI that runs where the data lives — locally, on a hospital's own servers or on edge devices like smartphones. 2,200+ Apache-2.0 models, 460M+ downloads on Hugging Face, 18M+ installs on PyPI, and 85 contributors so far — with zero paid marketing.

Alongside OpenMed, I work in post-training at Arcee AI — developing open-weight models in the U.S. and releasing them for everyone — and I run AI & HPC infrastructure at CNRS.

My stance hasn't changed: medical AI must be open, auditable, and sovereign — deployable inside a hospital's own walls, and increasingly on the phone in your pocket. Its limits belong in code: OpenMed ships refusal and review gates, not terms-of-service promises.

And it must be measured in the open: what ran, what it can do, where it failed, and what it still cannot claim. That is the standard I hold my own models to, and the one I'd hold any health model to.

Ask AI about Maziyar Panahi

Dig into my work in health AI, the OpenMed story, or open medical models.

ChatGPT Claude Perplexity Grok

OpenMed

Open-source clinical AI: 2,200+ Apache-2.0 models for de-identification, clinical NER, coding, and QA — built to run locally — on-prem, in your VPC, or on edge devices like smartphones. 460M+ downloads on Hugging Face, 5,000+ GitHub stars, and the #1 most-referenced organization in Hugging Face's State of Open Source report (Spring 2026).

  • openmed on GitHub — Python package, CLI, and on-device runtimes; near-weekly releases, 85 contributors
  • OpenMed on Hugging Face — the model hub
  • OpenMed NER — state of the art on 10 of 12 public biomedical NER datasets
  • OpenMed releases — multilingual PII across 42 language routes (35 model-backed), ~51 offline ID validators, and a k-anonymity release-risk workflow that refuses to auto-sign expert determinations

Next: OpenMed's own reasoning models — on-device, local-first — and multimodality across X-ray, pathology, and medical speech-to-text.

OpenMed Agent

A clinical AI agent built on OpenMed's open models — terminal-native, inspectable, privacy-first, reviewer-gated, and running on-device on smartphones. Early access is open at agent.openmed.life.

Welna

Know your numbers. Own your health story. Welna is my consumer iOS app, powered by OpenMed: it reads your Apple Health data, redacts identifiers on-device, and preps you for your next doctor's visit — your name, dates, and IDs never leave the phone.

Posts

Real device, real input, visible result — demos over decks.

Research

Models

600+ of these run natively on-device, on laptops and smartphones.

RL Envs

51 open medical RL environments on Prime Intellect — the same stack used to post-train Arcee's Trinity Mini for scientific tool-use.

Show all 51 environments

About

I've spent 16 years in public academia: twelve at CNRS, where I'm a lifetime civil servant and run the AI & HPC infrastructure at ISC-PIF (140+ servers, 380B+ documents), after research-assistant and cloud-architect years at Multimedia University and the University of Malaya.

Before OpenMed, from 2019 to 2025, I was at John Snow Labs leading Spark NLP — 150M+ downloads and 100,000+ pretrained models and pipelines. Since 2024 I've also been in post-training at Arcee AI: GRPO, RL environments, and the INTELLECT-1 and Trinity technical reports.

What hasn't moved: medical AI should be open, auditable, and able to run inside a hospital's own walls — with the models, the training, and their limits out in the open. The longer version is on LinkedIn.