Confluence Labs
Confluence Labs is an AI research lab focused on learning efficiency, helping researchers and engineers make breakthroughs in data-sparse domains like drug design and materials science with the fewest possible experiments. The company combines LLMs with discrete program search to generate hypotheses and build data-efficient models, achieving state-of-the-art 97.9% accuracy on the ARC-AGI-2 benchmark at roughly $12 per task.
- Artificial Intelligence
- Drug Discovery & Therapeutics
- Software Only
- New Materials
- Biotechnology
Funding
Raised to date
$500KRaised to date based on public sources. This may differ from the amount the company actually raised and is based only on what is publicly available on the internet.
Founders
Founder details are not available yet.
Product
Problem
Many of humanity's biggest advances, such as the right drugs to end human disease or the right materials to unlock fusion, are locked behind a certain number of physical experiments. These constrained search problems are expensive and time-consuming, and modern machine learning struggles in data-sparse domains where experimental data is costly or difficult to obtain.
Solution
Confluence Labs is an AI research lab focused on learning efficiency—how quickly a system can learn given limited data or the fewest number of experiments. The company uses LLMs combined with discrete program search to generate hypotheses and build data-efficient models that accelerate frontier research. Its approach structures problems to resemble LLM training data, enables long-horizon reasoning, and precisely measures solution criteria to find optimal solutions with minimal experimental cost.
Target Audience
Primary collaborators and customers are researchers and engineers in hardware engineering, biology, materials science, and other fields where progress is gated by the cost of physical experiments.
Features
- LLM-driven program synthesis that writes code describing transformations in ARC-style problems, enabling reasoning from just a few examples
- A hypothesis generation capability that designs maximally informative experiments to reduce development time for new technologies
- Data-efficient modeling that combines LLMs with discrete program search for accurate predictions in data-sparse domains such as hardware engineering and drug design
- State-of-the-art performance on ARC-AGI-2 with a 97.9% score at roughly $12 per task on the public evaluation set
- Open-sourced solver available for reproduction on GitHub