Built by engineers who felt the pain
Aryan and Lena spent years inside the infrastructure that powers today's largest AI deployments — at DeepMind and Scale AI — before founding Vektor in 2023. The company is the answer to a question they kept hitting from both sides: why is it still this hard to know whether an agent is actually working in production?
Aryan Mehta
CEO & Co-founder
Spent four years at DeepMind as a Staff Engineer, building the eval harnesses that quietly catch an RL policy gaming its reward before anyone ships it. Walked away from a Stanford CS PhD halfway through to co-found a developer-tooling startup that Stripe acquired in 2021. Has shipped to production with reinforcement-learning agents that most teams only read about.
Lena Kovacs
CTO & Co-founder
Owned the training-data evaluation pipeline at Scale AI as a Principal Engineer, then went on to design distributed tracing for two of the largest LLM providers running today. MSc from ETH Zürich, with a thesis on streaming observability for high-throughput inference systems. Believes the boring infrastructure problems are where agents quietly die in production.
The team behind the founders
Vektor is 18 people across infrastructure, ML, product, and design — a small group that moves fast and cares deeply about craft. What we're building is bigger than the founders' stories: an opinionated platform for evaluating, debugging, and shipping AI agents that behave the same in production as they do on your laptop. If any of that resonates, the broader Vektor story is worth a look.