Project
QDevBot — Scientific Experiment Agent
Project Brief
A high-level scientific agent for long-running experimental work. It carries planning, retrieval, memory, and iteration across runs while delegating instrument execution and scan assessment to lower layers of the control stack.
Project Tech Stack
Approach
The runtime keeps scientific intent separate from raw device state, carrying planning, retrieval, and memory across long-horizon runs instead of forcing every operational detail into the same context window.
It coordinates lower layers such as instrument execution services, STM scan-quality assessment, and benchmark feedback so the agent can revise plans with structured context rather than direct device access.
That makes QDevBot the cognition layer of the autonomous experiment ecosystem: it decides what to try next, uses prior context to iterate, and delegates bounded device-facing work to specialized services.
Outcome
QDevBot provides high-level experimental reasoning for the control stack while keeping execution, perception, and recovery responsibilities separated into inspectable lower layers.
Media
