AI Agent for Liquid Handling Instrument
Liquid handling robots play a crucial role in life sciences laboratories, but their control software lags behind the advancements in modern AI. Scientists often spend countless hours programming pipetting sequences manually through inflexible interfaces specific to vendors, even for standard experiments. Additionally, non-technical lab personnel require extensive training to operate these instruments. Currently, there is no standardized method for integrating AI with physical devices while meeting the safety requirements demanded in regulated environments, such as schema validation, human oversight, and comprehensive audit trails.
To address this issue, we developed an AI platform that operates on liquid handling instruments, allowing for natural language control instead of manual programming. Scientists can describe their experiments in simple English, and the AI agent takes care of planning the workflow, validating each parameter against hardware limitations, presenting the plan for human approval, and executing it on the robot. The system features a three-service architecture, which includes a React frontend for real-time 3D deck visualization, a FastAPI backend for managing instrument states and tracking labware, and a CrewAI agent server connected through the Model Context Protocol (MCP), an open standard for communication between AI and tools. A planner-executor pattern ensures that the AI only sends validated commands to the hardware, and every action is logged for complete traceability.