Erik Gafni is a cofounder and CTO of TypeSafe AI. His earlier machine-learning work applied probabilistic models and neural networks to genetic diagnostics and blood-based cancer screening.[1][2]
Start with COSMOS: Python library for massively parallel workflows →
GitHub 
X Career
Clinical genetics and cancer screening
Gafni describes work at Invitae in 2013–2017 as a senior bioinformatician developing genetic-diagnostic algorithms. At Freenome in 2017–2018, he developed neural-network and probabilistic models for colon cancer detection.[2]
Contributions
Learning from DNA sequences
At Ravel in 2018–2021, Gafni applied architectures from language, vision and audio to sequencing data for detecting breast cancer in cell-free DNA.[2]
Managing large research workflows
Gafni is a joint first author of COSMOS, a Python library for describing and executing computational pipelines. It separates a tool’s expected input and output types from specific filenames, then builds a dependency graph of jobs. A workflow can run across a computing cluster or cloud resources while a user interface tracks progress. This addresses a common genomics problem: coordinating many analysis steps across large numbers of samples without manually managing every job.[3]
Tools, writing & talks
Writing, research & talks
Start here
COSMOS: Python library for massively parallel workflowspmc.ncbi.nlm.nih.govA workflow library that separates analysis tools from job dependencies and schedules pipelines across distributed resources.
Start with the paper’s workflow diagram to see how separate analysis steps become a parallel pipeline. Some Python and familiarity with batch computing help; genomics expertise is not needed to understand the scheduling design.
Joint first author[3]