Scaling Quantum Materials Research:

A high-level scientific agent runtime for long-horizon experimental work. It carries planning, retrieval, memory, and iteration across runs without absorbing raw device uncertainty into the same context, and coordinates lower layers to decide what to try next.

A bounded execution service for live instrument work under uncertain conditions, where device state is only partially observable and execution can drift. It enforces an action-observation-update loop with state verification and exposes that control layer over HTTP to higher-level agents.

A driver framework that turns heterogeneous instrument APIs into a consistent control surface. It provides machine-readable capabilities and reusable control contracts so new hardware can be added without bespoke orchestration logic for each device.

A perception service that evaluates STM scans from images and scan metadata. It turns raw scan output into compact quality judgments that higher layers can use without directly consuming measurement-level detail.

A task-construction pipeline that turns quantum-physics papers and datasets into benchmark-ready tasks, task profiles, scoring inputs, and provenance-tracked exports. It supplies the evaluation stack with structured scientific work rather than isolated question-answer pairs.

A benchmark harness for evaluating multi-step agent behavior in quantum research environments. It traces tool use, intermediate decisions, grading signals, and task trajectories rather than relying only on final answers.