Iluvatar Labs
Iluvatar Labs is an AI-native research lab building closed-loop autonomous systems for post-training, scientific research, and bioinformatics. Its stated aim is to close the loop between model training and downstream behavior through continuous monitoring, automatic adjustment, and rich context tracking.
At a Glance
- Academic researchers and research laboratories
- AI and machine-learning teams
- Computational biology, bioinformatics, and life-science organizations
- Enterprise R&D teams
- +2 more
AI Tools by Iluvatar Labs
(1)SITREP
Self-Hosted Claude Code Dashboard
Discussions
No discussions yet
Be the first to start a discussion about Iluvatar Labs
Latest News
Introducing Maiar-27B, a fast scientific-evidence decision model built from Valar with Kev
Introducing Valar-27B, a scientific reasoning model for long-running investigations
Randomized Gradient Compression research announcement
Marvin Atlas launched with 250M+ scientific records and text from 25M papers
Products & Services
An autonomous AI co-scientist/research agent that performs literature review, hypothesis generation, experiment design, execution, analysis, and documentation across ML, computational biology, materials science, and other disciplines. It maintains a logic trail and persistent project context, and can operate at user-selected levels of autonomy.
A patent-pending closed-loop control layer for model transformation. It monitors protected metrics, automatically adjusts training pressure when capabilities drift, applies guardrails, and certifies runs with an audit trail across fine-tuning, alignment, distillation, and compression.
An open research program powered by Marvin that invites people to propose questions, scrutinize analyses, and contribute evidence. Initial programs focus on sarcopenia and schizophrenia; publications, code, and analyses are intended to be open source.
A scientific literature search capability inside Marvin covering more than 250 million records, including papers, datasets, clinical trials, and regulatory documents, with searchable text from 25 million papers down to passages in Methods and Results.
Market Position
Iluvatar Labs positions Marvin as a rigorous, human-supervised autonomous-science collaborator rather than a simple chat or coding assistant, emphasizing persistent context, provenance, logic trails, domain-expert scientific skills, and end-to-end research loops. Its own benchmark discussion compares Marvin with Biomni, Claude Code, and other agentic platforms. Actuator is positioned against fragmented, open-loop post-training stacks and post-hoc evaluation by providing continuous monitoring, training-time correction, guardrails, and certification.
Leadership
Founders
Ben Q. (Yuan) Zhao
Founder and CEO of Iluvatar Labs. He holds an SB in Biological Engineering from MIT and a PhD in Bioinformatics and Systems Biology from UC San Diego. Before Iluvatar Labs, he was Head of AI & ML and Senior Biological Data Scientist at Encoded Therapeutics, a graduate research assistant at Ludwig Cancer Research, and a visiting research fellow at Japan's National Institute of Advanced Industrial Science and Technology (AIST).
Executive Team
Ben Q. (Yuan) Zhao
Founder and CEO
Machine-learning and computational-biology researcher; previously Head of AI & ML/Senior Biological Data Scientist at Encoded Therapeutics, with research experience at Ludwig Cancer Research and AIST.
Founding Story
Iluvatar Labs says its core thesis is that much of today's AI stack is open-loop: models are trained, evaluated, and deployed without continuous correction. The lab was started to build closed-loop autonomous systems that monitor behavior, adjust training or research workflows, preserve context, and keep humans over the loop where judgment matters; Marvin was built because existing research tools were too error-prone, wasteful, or opaque for the team's own work.
Business Model
Revenue Model
Marvin is offered as a free trial/product with enterprise access, on-premises deployment, security, and other enterprise requirements handled through a sales/demo process. Actuator is in closed testing/closed beta with early design partners and potential pilots or partnerships. Publicly released projects such as SITREP, Kev, and AgentPad13 are open source.
Pricing Tiers
The Marvin site offers a 'Try Marvin Free' path; the page does not state quantitative limits in the fetched material.
Enterprise access, on-premises deployment, security, and other requirements are handled through a demo/contact process.
Target Markets
- Academic researchers and research laboratories
- AI and machine-learning teams
- Computational biology, bioinformatics, and life-science organizations
- Enterprise R&D teams
- Organizations training, aligning, distilling, or compressing AI models
- Individual developers and teams running Claude Code, Codex, or Hermes agents
- Autonomous and assisted scientific research
- Machine-learning and AI research
- Computational biology, bioinformatics, and gene-therapy research
- Literature review and evidence synthesis
- Hypothesis generation and experimental design
- Paper reproduction and scientific-data analysis
- Stanford University
- Massachusetts Institute of Technology
- Broad Institute
- Washington University in St. Louis