Heavybit Welcomes New Member: Vibrant Labs

Vibrant Labs opens a new frontier in AI infrastructure: production-grade, RL-ready simulation and verifier-driven evaluation built for long-horizon agents. By building complete, RL-ready worlds and developer-first workflows, Vibrant gives agents safe places to train, fail, learn, and improve before they ever touch production. Making evaluation as natural as testing, Vibrant provides developers the confidence to ship new AI features. Heavybit is thrilled to back a team tackling one of the biggest challenges in AI development today.” -Jesse Robbins, General Partner/Heavybit

Heavybit is thrilled to announce our investment in Vibrant Labs. Vibrant is constructing the simulation environments and streamlined developer workflows necessary to safely evaluate, train, and iterate on long-horizon agents. This work represents the foundational next layer for any team that relies on AI agents for complex reasoning, persistent action, and continuous self-improvement.

Why This Matters: Long-Horizon Agents Are the New Frontier

AI is no longer only about single-turn interactions or answering questions. Teams are building systems that must plan, coordinate, and act across many steps, what we call long-horizon agents. These agents look less like calculators and more like operators: They open a support case and carry it to resolution, shepherd a release from first commit to production, or manage logistics and operations that unfold over days.

That new capability creates a single clear problem: Current engineering practice lacks a safe, repeatable, production-grade place to measure, train, and improve agents whose failures and successes play out over long periods and many steps. Benchmarks and unit tests catch surface errors; they miss emergent, multi-step failure modes. Training models in production is dangerous. Vibrant Labs answers that gap by building the simulation infrastructure and developer workflows teams actually need.

Ragas.io to Vibrant

Two years ago, Shahul ES and Jithin James started with a focused thesis: If teams make evaluation part of everyday development, they’ll catch mistakes during development, not in production. They released Ragas.io, the leading open-source evaluation and verification framework. Ragas has proven two essential things at scale:

• Evaluation can and should be everyday engineering work.
• Verifiers and evaluation artifacts are not only checks. They can be training signals.

Vibrant Labs is the natural continuation of that journey: Take the hard lessons from shipped evaluator tooling and build complete, instrumented, RL-ready environments where agents can practice, fail safely, and improve before they touch live systems. The founders have been actively iterating product and docs through public and private workspaces as they transformed Ragas into a platform for environments and RL training.

Why We’re Excited

Vibrant Labs is built on state-of-the-art research and the incredibly popular Ragas.io open source evaluation and verification framework, which has more than 10k stars and significant institutional adoption. The Vibrant Labs team has proven that evaluation can be made part of day-to-day development, and verifiers can become a signal for training. They’re building the operational layer teams will depend on:

• Production-grade, RL-ready simulation worlds: Rich, instrumented environments that model the complexity of long-horizon tasks so agents can explore, practice, and generalize.
• Verifier-driven evaluation: Integrated verifiers that explain why behaviors fail, producing explanatory signals teams can use to debug and improve agents.
• Developer-first workflows: Tooling that makes evaluation as natural as testing — repeatable, auditable, and part of the developer lifecycle.
• Training pipelines that combine verifier feedback with RL: New training paradigms that convert evaluation artifacts into supervision so agents can learn faster in simulation.

Put simply: Vibrant closes the loop from evaluation → training → validation → deployment in a way that’s safe and repeatable. We’re excited to partner with Shahul and Jithin as they build the next layer of AI infrastructure — the RL-ready, verified environments teams need to move agents from experiments into reliable, production work.

Meet the Founders

Shahul Elavakkattil Shereef

Co-Founder

Shahul Elavakkattil Shereef is one of the co-founders of Vibrant Labs and the Ragas.io project. A veteran data scientist and machine learning engineer, he has served tours of duty at Kaggle, Neptune, and AMPLYFI.

Jithin James

Co-Founder

Jithin James is one of the co-founders of Vibrant Labs and the maintainer of the Ragas.io project. He has previously served in ML engineering roles at BentoML and Trell.