The research engine
Ideas. Experiments.
Evidence.
Gnosys coordinates specialist research agents to explore competing ideas, run experiments, and develop models. Every result feeds the next round of research.
How it works
Start with a question
Define the outcome, baseline, and budget. Identify the evidence needed to judge a result.
Coordinate competing approaches
Agents explore models, implementation, and features. They share results and review what to try next.
Develop useful measurements
Agents build and revise the measurements that guide their search. We call this objective engineering: developing better feedback while keeping the final goal fixed.
Evaluate the selected result
Reproduce the selected candidate and evaluate it separately. Report the comparison, failures, costs, and limits.
Current work
From model improvements
to better research methods.
Our autonomous ML runs and coordinated-agent experiments span search, reproduction, and final evaluation. Our CausalML study measures how objective engineering changes the research team’s results.
The next step is recursive improvement: an improved researcher takes over the next round and develops its successor. We’re designing experiments to measure that progress across generations.