Skill Doctor: Autonomous diagnostics for model tooling and observability
A diagnostic design for checking agent tool contracts, permissions, dependencies, and live availability before failures spread through long workflows.
Recursive context accumulation in autonomous scientific environments
How papers, repositories, execution traces, and failed experiments become versioned evidence that later agent runs can retrieve and verify.
Kalaris Research
Deterministic evaluation benchmarks for autonomous multi-model scientific systems
A benchmark protocol for separating model regressions from changing tools, environments, and datasets in long-horizon scientific agent workflows.
Edge delivery without architectural fragmentation
An architecture note on serving prerendered research content and a small number of request-time operations through one observable deployment boundary.
Topological graph routing for distributed agent clusters in chemical synthesis
A context-routing design that gives retrosynthesis, property, evidence, and safety agents only the graph regions needed for a planning decision.