About VAIL
VAIL was founded in 2025 to address the need for transparency and verification of AI systems.
Team
VAIL is built by entrepreneurs, engineers, and researchers who share a belief that AI systems should be transparent, verifiable, and trustworthy. We combine expertise in data science, machine learning, cryptography, and systems engineering to build the tools and infrastructure for safe and secure AI systems.
-
Manish Shah
Co-founder & CEO
Manish is a repeat co-founder, technology enthusiast, and has been building companies for 20+ years with a passion for products and systems that help people live better lives. Previously he was co-founder of LiveRamp ($RAMP), a data platform company, and PeerWell, an AI-powered digital health service. Manish holds a degree in Computer Science from UC Berkeley.
-
Jonah Leshin
Co-founder & CTO
Jonah is a repeat co-founder and technical leader. He previously led machine learning and product teams at Datavant focused on privacy-preserving technology for healthcare data. Jonah is a published author in medical informatics, AI systems, and theoretical mathematics and holds Math degrees from Northwestern University, University of Cambridge (Master’s), and Brown University (PhD).
-
Joel Jewitt
Chief Business Officer
Joel is a technology operator and executive for cutting edge companies over several decades. He was on the founding team of Palm Computing, Good Technology, and LiveRamp. Joel has a degree in Physics from Amherst College and an MBA from Stanford University.
-
Daniel Kang
Founding Technical Advisor
Daniel is an assistant professor at UIUC in the computer science department. He was a postdoc at UC Berkeley in the Sky lab, working with Ion Stoica, and a PhD student at Stanford working with Peter Bailis and Matei Zaharia. Daniel also helped develop DawnBench and MLPerf. His research focuses on verifiable computation for machine learning, adversarial robustness, and ML benchmarking. Daniel has a degree in Computer Science from MIT, a Master’s from University of Cambridge, and a PhD from Stanford University.
-
Ian Timmis
ML Platform Engineer
Ian has researched and developed deep learning methods for use-cases in natural language processing and computer vision. Applications include pose recognition, motion conformance, and range of motion measurements. He’s deployed models to production from data centers to mobile devices. He has a Master’s in Computer Science from Lawrence Technological University.
-
Megan Mehta
Model Behavior Scientist
Megan studies how AI models behave in practice, bringing a data science lens to model evaluation, transparency, and trust. She holds a B.A. in Data Science and a B.S. in Society & Environment from UC Berkeley, where her work spanned natural language processing, sentiment analysis, and applied machine learning. She combines technical and interdisciplinary perspectives to help make AI systems more understandable and accountable.
Company principles
- Transparency
- Tools for implementing verifiable AI should be open source. Model metadata that attests to authenticity, safety, and performance on various benchmarks should be open. Verifiable AI tools should also be compatible with models and training methods that are not open source due to competitive or security concerns.
- Verifiability
- Assurances related to model training and execution should be technologically demonstrable, for example via cryptographic guarantees and hardware-level attestation. Contractual guarantees of model usage are critical for building trust, but they are not sufficient on their own, given the complexity of AI systems and the possibility for misunderstanding.
- Community stewardship
- AI developers should be able to post proof of their own model training and usage, and verify proof of others. Verifiability tools should be easy to adopt for honest actors, in technical barrier to entry, computational cost, and compatibility with AI development, and difficult for bad actors to abuse. Proofs, verifications, and model metadata should serve as a tool that the AI community, policy makers, and regulators can use to inform decisions on AI development and governance.
- Standardization
- There should be clear standards for how information related to verifiability is generated, structured, and made available. These standards should meet the needs of both researchers and practitioners.
Join us
We’re looking for people who share our vision of transparent and verifiable AI. Whether you’re a researcher, engineer, designer, or domain expert, there may be a place for you on the team.