Veris AI Raises $8.5M to Bring “AI Gyms” to Life

Veris AI is building for a critical layer of the stack: training infrastructure. The New York-based startup, which recently emerged from stealth with an $8.5 million seed round led by Decibel Ventures and Acrew Capital, believes the barrier to enterprise AI is the environments in which agents are trained.

Veris AI provides high-fidelity simulation environments where AI agents can learn, practice, and improve, much like self-driving cars tested in virtual cities before hitting the road. Veris is expecting these agents, trained not with static prompts but through interactive experience, to form the backbone of enterprise automation in the coming years. “We started Veris after observing that the traditional ways we train machine learning models just don’t work for AI agents,” said Mehdi Jamei, Veris AI’s co-founder and CEO. “Similar to how self-driving cars needed simulated cities to become production-ready, AI agents need real experience.”

Jamei brings both technical depth and enterprise AI experience to the table, having led agentic AI initiatives at System and Workmate. His co-founder and CTO, Andi Partovi, previously worked as a Solutions Architect at Google and co-founded KeyLead Health. Together, they combine academic credentials: a Ph.D. in electrical engineering and brain-computer interfaces respectively, with years of building applied AI systems for business use.

“As founders, Mehdi and Andi have a rare combination of academic, technical and enterprise customer experience that makes them well-positioned to deliver on this new paradigm,” said Alessio Fanelli, Partner and CTO at Decibel Ventures. “We’ve been looking to make an investment in this space for quite some time and Veris is the only platform we found that is building the environment layer that enterprise AI has been missing.”

A Simulated AI Agent Training Ground

Veris AI is building the training grounds where agents learn to behave in ways companies can trust. These environments allow agents to interact with realistic data and systems, make decisions, and receive feedback through reinforcement learning or fine-tuning methods.

Curretnly working with an early group of clients, they cite the following use cases: a consumer fintech firm uses Veris to simulate user conversations, disclosures of sensitive information, and potential regulatory violations. The goal: ensure chatbots are both compliant and capable before they’re deployed. Similarly, an HR tech company is using the platform to teach AI executive assistants how to manage calendars and handle confidential communications reliably and securely. A manufacturing firm, meanwhile, is training agents to handle procurement scenarios, including supplier research and risk evaluation.

“The underlying brain of an agent is an LLM, and there are some ways of improving it over time, but it’s generally memoryless from call to call, and that’s another aspect of the training ground that we’re building: to be able to go back and optimize and fine-tune the agents in some shape or form,” said Partovi.

These simulations serve not only as sandboxes for agent behavior but also as feedback loops for ongoing performance improvement. “These are ripe for fine-tuning, because it turns out they may be quite bad after fine-tuning on a lot of other things, like they’ll be terrible at poetry, probably, but really good at that thing that they have to do, which is what enterprises actually care about,” said Jamei.

The concept of training AI agents through simulated experience is still relatively new, but competition is emerging. Some companies attempt to manage agent development using prompt engineering or human-labeled datasets. Others, like Adept or Imbue, are exploring broader frameworks for general-purpose agents. But Veris is focused narrowly on infrastructure.

“Veris is ahead of the curve in approaching agent training through experience rather than evaluation,” said Asad Khaliq, Founding Partner at Acrew Capital. “Their simulation-first approach is exactly what enterprises need to confidently deploy AI agents in production. We believe Veris is defining a new and dynamic category in enterprise AI, and expect the rest of the industry to quickly follow their lead.”

Connecting AI Agents to Enterprise Use Cases

While foundation models continue to improve, most enterprises still struggle to integrate autonomous agents into production systems. Veris aims to close that gap. “Today, most agents lack the accuracy, consistency and governance required for enterprise use, exposing organizations to significant risk,” Jamei said. “Veris exists to eliminate these roadblocks by allowing developers to train agents using experience rather than prompt engineering and human-generated data.”

With its early customers spanning fintech, manufacturing, and enterprise SaaS, Veris is signaling that simulated agent training is in active use today. If the company succeeds in establishing itself as the de facto training layer for AI agents, it could become an essential part of the agentic stack.

According to Jamei: “We are building Veris to unlock the potential of agentic AI for enterprises—both by solving existing problems and improving the speed and quality in which new agents can come into production.”