The puppeteer of experiments
Geppetto is a cloud platform for AI-driven laboratory science. Design protocols visually or in plain language, run them on automated hardware, and let AI analyze each round and propose the next — closing the Design–Build–Test–Learn loop across cloud and bench.
Geppetto is in beta and accounts require an invite code — email adam.gormley@rutgers.edu to request one.
A campaign is many iterations. Geppetto carries each one from idea to result and back to idea.
Design
Build liquid-handling protocols on a drag-and-drop canvas, or describe the experiment in plain language and let the AI assistant write the steps. Everything is validated against your labware as you work.
Transfer · CORE96 · HeaterShaker · iSWAP · UVVisReadBuild
Experiments are assembled from capability sets your lab publishes — pre-validated combinations of instrument, labware, and method. Invalid configurations are impossible to express, so errors are caught before anything reaches the bench.
CLAP · Cloud Lab Access ProtocolTest
Submit to automated hardware — Hamilton liquid handlers running PolyCraft — or run the protocol by hand and upload results from any lab. An operator approves every run before it starts.
Hamilton · PolyCraft · manual uploadLearn
AI analysis turns each iteration's raw data into metrics, insights, and a proposed next experiment. You review the proposal, and the loop begins again.
per-iteration analysis · protocol diffs · next-experiment proposalsThe loop closes
Undo, redo, keyboard shortcuts, and step-by-step editing on an interactive canvas.
A chat assistant that adds, edits, and reorders steps as you describe what you need.
Steps are checked against deck layout and labware constraints as you build.
A per-iteration timeline of analyses, plots, protocol diffs, and project files.
With robots
Labs with Hamilton systems connect through CLAP nodes: protocols travel from the cloud to the instrument, and results flow back automatically.
Without robots
No automation hardware? Run the protocol at your own bench and upload the results. The AI analysis and the loop work exactly the same.
A human stays in the loop at every step: no protocol runs without operator approval, and no proposed experiment is submitted without your review.