Ali Madani

Research

At Salesforce AI Research, Madani was the architect of ProGen, which posed protein engineering as an unsupervised sequence generation problem. The 2020 preprint, co-authored with Richard Socher and others, trained a 1.2B-parameter model on about 280 million protein sequences with tags such as molecular function and cellular component.³¹

The follow-up paper, published in Nature Biotechnology in January 2023 with Madani as first author and affiliations at Salesforce Research and Profluent Bio, reported that large language models generate functional protein sequences across diverse families.²

Profluent

As Profluent's co-founder and CEO, Madani leads a team combining machine learning and biology to design proteins, with the platform first pointed at CRISPR and gene editing. Under its OpenCRISPR initiative the company released OpenCRISPR-1, an AI-created gene editor that keeps a Type II Cas9 architecture while differing from SpCas9 by more than 400 mutations.¹⁴

In November 2025 Profluent announced Profluent-E1, retrieval-augmented protein encoder models trained on its Profluent Protein Atlas and released in three sizes for research and commercial use.⁵

Work and contributions

• ProGen Published research

  First author of the 2020 ProGen preprint, which trained a 1.2B-parameter language model on about 280 million protein sequences conditioned on taxonomic and keyword tags, and of the 2023 Nature Biotechnology paper showing that such models generate functional proteins across diverse families.³²

• OpenCRISPR-1 Published research

  Leads the company behind OpenCRISPR-1, an AI-generated gene editor that Profluent released under its OpenCRISPR initiative, freely licensed for ethical research and commercial use.⁴¹

• Profluent-E1 Open weights

  Leads the company that released Profluent-E1, a family of retrieval-augmented protein encoder models in 150M, 300M and 600M parameter sizes, under a permissive license.⁵¹

• Profluent-E1 150M (E1-150m) Open weights

  Last-listed and corresponding author of the E1 preprint (bioRxiv 2025.11.12.688125), which the Profluent-E1 repository README cites for the model family.⁶⁷⁵

• Profluent-E1 300M (E1-300m) Open weights

  Last-listed and corresponding author of the E1 preprint (bioRxiv 2025.11.12.688125), which the Profluent-E1 repository README cites for the model family.⁶⁷⁵

• Profluent-E1 600M (E1-600m) Open weights

  Last-listed and corresponding author of the E1 preprint (bioRxiv 2025.11.12.688125), which the Profluent-E1 repository README cites for the model family.⁶⁷⁵

Timeline

1. Mar 8, 2020Submitted the ProGen preprint on language modeling for protein generation.³
2. Jan 26, 2023Published, as first author, the ProGen paper in Nature Biotechnology.²
3. Nov 13, 2025Profluent announced the Profluent-E1 retrieval-augmented protein encoder models.⁵

Other documented roles

Roles the sources document that are not confirmed as current.

• Researcher; architect of the ProGen project, Salesforce AI Research¹³

Where they work

• Profluent 3 people in the wiki

Related people

• Sarthak JainAlso at Profluent
• Jeffrey A. RuffoloAlso at Profluent
• David WaddenLanguage models · AI for science
• Alex RivesAI for science · Language models
• Liam FedusLanguage models · AI for science
• Jakob UszkoreitLanguage models · AI for science
• Kyunghyun ChoLanguage models · AI for science
• Yejin ChoiLanguage models · AI for science
• Garyk BrixiAI for science · Language models
• Hoifung PoonAI for science · Language models