Coscientist
Autonomous chemical research system that generates liquid-handler control scripts and runs palladium-catalyzed cross-couplings, with its condition optimization run against previously collected, fully mapped yield datasets.
The loop
- Proposes
An LLM planner designs reaction matrices and writes Opentrons Python protocol scripts and hardware commands.
- Judged by
Protocol syntax checkers, robotic liquid handlers, and UV-vis plate readers execute the reactions and measure product yield.
- Steers
The optimization study selects conditions against previously collected, fully mapped yield datasets.
Headline result
Autonomously executed Suzuki-Miyaura and Sonogashira cross-coupling reactions on cloud lab hardware and ran an optimization study that selected conditions against previously collected, fully mapped yield datasets.
What this does not prove
Wet-lab loop throughput is bounded by physical incubation and analytical characterization times.
How it generalizes the harness
Physical robotic reactors and spectrophotometric yield measurements judge the proposed experimental protocols.
Classification
| What it proposes | Program |
|---|---|
| Generative model | GPT-4 experimental planner and protocol synthesizer |
| Algorithmic authority | Opentrons robotic execution validator and UV-vis spectrophotometer |
| Feedback | Score |
| What gets updated | Prompt context |
| When it runs | In continuous operation |
| Validity | Approximate physics (simulation, not lab results) |
| Evidence | Experiments reported |
| Lineage | Original system |