Skip to main content

01 About

01

Founder of VerySafe.ai

AI Safety Researcher & Engineer · Vice Chair, MLCommons Medical Working Group

AI researcher, engineer, and founder building trustworthy, end-to-end AI systems for regulated, high-stakes domains. I hold a Ph.D. in Computer Science from the Technical University of Munich (summa cum laude) and bring 11+ years of operationalizing AI from prototype to production.

I am the founder of VerySafe.ai, where I am building SafeCompute — a policy-aware compute platform that attaches cryptographic proof to every AI model run through remote attestation, supply-chain provenance, and signed audit lineage. The aim is to let organizations run frontier and open-source LLMs precisely where governance, privacy, and auditability are non-negotiable.

My work sits at the intersection of frontier AI capability and the safety, attestation, and governance infrastructure required to deploy it responsibly. I have led USD 9M+ in NIH/NCI-funded research, published in Nature, Nature Communications, and IEEE Transactions on Medical Imaging, and serve as Vice Chair for Algorithmic Development at the MLCommons Medical Working Group.

I believe open software fosters better science, which is why I stay deeply involved in the open-source community.

02 What I build

From research to production

I architect and ship AI systems end-to-end — concept, design, deployment — for regulated, high-stakes domains. Eleven years turning research into software that runs in production.

End-to-end AI systems

Architect and ship AI from prototype to production — multimodal data, low-code pipelines, clinical-grade workflows.

GaNDLF — 30% faster prototyping, now an MLCommons project

Confidential & federated compute

Privacy-preserving ML that trains and benchmarks across institutions without moving sensitive data.

USD 9M+ in NIH/NCI grants led · deployed on 6 continents

Optimization & deployment

Make models run where compute, latency, and cost are constrained — edge, HPC, and low-resource environments.

10–50% less compute · up to 70% lower inference latency

Benchmarking & evaluation

Trustworthy, reproducible evaluation of medical and enterprise AI at scale.

MedPerf — federated benchmarking across institutions

Got a problem that needs shipping?

Let's talk — discovery call on me.

sarthak@verysafe.ai