The new platform offers a lifelong learning engine that turns real-world failures into verified improvements, making agents more reliable the more they are used.
BETHESDA, MD / ACCESS Newswire / June 10, 2026 / RELAI today launched a verifiable continual learning platform for AI agents, and announced $6.9 million in total funding to scale it. The funding includes a newly secured $5.4 million pre-seed round led by .406 Ventures with participation from AITFund (“AI Tinkerers Fund”) and other strategic investors, along with $1.5 million in prior investment support from Non sibi Ventures and TEDCO. RELAI will use the capital to expand its engineering team, further develop the platform, and prepare for broader go-to-market efforts.
As enterprises move AI agents into production, keeping them reliable after deployment has become one of the hardest unsolved problems. Agents fail unpredictably, and fixes often create silent regressions, leaving teams stuck in a cycle of prompt patches, rerun evals, and reactive debugging.
The primary issue is that learning is rarely verified against what already works. RELAI solves this by turning failures, traces, evaluations, and human feedback into replayable learning environments, where each failure becomes a reusable signal for durable, verified improvement.
The company was founded by Soheil Feizi, an associate professor of computer science at the University of Maryland whose research focuses on AI reliability and failure analysis. He earned his PhD from MIT and, in 2025, received the Presidential Early Career Award for Scientists and Engineers (PECASE), the highest honor the U.S. government grants to early-career scientists and engineers. Feizi’s research group and collaborators have produced more than 100 AI research papers and amassed over 15,000 citations.
“Getting an AI agent into production is no longer the hardest part; keeping it reliable as teams continuously improve it is,” said Kevin Wang, Principal at .406 Ventures. “Soheil has spent his career studying how AI systems fail, and RELAI turns that research into practical infrastructure that helps enterprise agents learn from failures without breaking what already works.”
Most systems that modify agents check for regressions after the change is shipped. RELAI takes a fundamentally different approach by keeping regression control inside the optimization loop. Every proposed improvement is continuously validated against a growing portfolio of prior environments as it is being searched, not after. The company calls this online, in-loop regression control, and it is the core of Verifiable Continual Learning – the way an agent gets better without becoming more fragile.
“At C3 AI, we’re delivering production-grade agents on the C3 Agentic AI Platform that take on complex, mission-critical workflows for enterprises across manufacturing, energy, defense, and more,” said Nikhil Krishnan, CTO & Chief AI Officer at C3 AI. “As these agents take on harder problems, the ability to evaluate and improve them on realistic edge cases becomes critical, and RELAI has helped us turn hard use cases into evals, and evals into measurable improvements in the agents we ship to customers.”
Just as important, RELAI routes each fix to the right layer of the agent stack. A failure might call for a prompt change, a tool wrapper, a memory update, a workflow adjustment, a model-routing decision, or a code-level repair. RELAI diagnoses the root cause and applies the smallest durable change at the layer where it belongs, instead of piling every fix into an ever-growing prompt.
In early deployments, RELAI lifted a financial services agent’s validation score from 39% to 80% and a healthcare proof-of-concept from 62% to 96% without the manual debugging loops those gains would normally require.
“For the past two years, the question was whether AI agents could use tools and pass benchmarks. They can,” said Soheil Feizi, Founder and Chief Science Officer of RELAI. “The real frontier now is whether agents can learn continuously from real experience without breaking what already worked. That is the gap RELAI is closing…the missing outer loop that turns failures into durable, verified improvement.”
RELAI’s continual learning engine also integrates with existing agent frameworks through a CLI and workflow integrations. It is designed to work alongside coding agents, orchestration tools, and enterprise AI stacks rather than replace them, so teams can enable Verifiable Continual Learning with just two commands. It also gives teams a persistent system of record for learning signals, optimization decisions, and regression history, so enterprises can see exactly how an agent’s performance evolves across deployments.
The company is backed by several major technology and research programs, including the NVIDIA Inception program, an SBIR award from the National Science Foundation, and an award from the Google Cloud for Startups program.
RELAI is opening limited-access onboarding today, ahead of a broader public release on June 22 and has already secured multiple customers and design partners. Early users will receive guided onboarding support and can join the waitlist at relai.ai.
ABOUT RELAI
RELAI is building continual learning infrastructure for AI agents. Its verifiable continual learning approach turns failures, traces, evaluations, and human feedback into replayable learning environments, identifies the root causes, and continuously optimizes prompts, tools, memory, workflows, and models with online, in-loop regression control. RELAI was founded by researchers and engineers specializing in AI robustness, evaluation, and failure analysis. Learn more at relai.ai.
MEDIA CONTACT
Nina Pfister of MAG PR, nina@mooringadvisorygroup.com
SOURCE: RELAI
View the original press release on ACCESS Newswire:
https://www.accessnewswire.com/newsroom/en/computers-technology-and-internet/relai-launches-verifiable-continual-learning-platform-for-ai-agen-1174434
