Is anyone exploring creating markdown files on Claude specific to liquid handlers development or scheduling programs? The thought popped into my head during my commute but would be interesting to see what it could do if you accumulate enough manuals or scripting guides. I’m sure the PyHamilton would be a great place to start to connect it, but for Beckman stuff the resources are pretty tightly kept.
Curious how you would also teach it to recognize physical constraints of the system such as deck positions/layouts, contamination concerns, or even hardware boundaries / motor limits.
Yeah thats a fantastic idea. I have been running small testing protocols directly in claude code so the basic functionality is there. I think the gap that markdown files can fill is like “teaching” claude code what different use patterns look like, for instance cherry picking, mixing beads, sample pooling etc. This is where a library of annotated protocols and unit operations can help a lot as examples for claude to reference.
Testing in simulation environments that can catch error cases like trying to reach a position thats out of bounds would be pretty important and can help the agent refine protocols.
I don’t expect manuals and guides to help much, they’re very detailed but in the wrong direction. Usually manuals teach a user about a GUI but Claude needs to know about the like physical capabilities and constraints of your platform and isn’t going to use the GUI at all.
Oh yeah that’s a pretty good point. Since pattern and syntax recognition is key to LLMs, I def agree that having a chest of well-documented methods for it to learn from would be another key approach.
And then over the weekend at my hackathon, I started using LabClaw on top of OpenClaw: https://labclaw-ai.github.io/
Physical constraints of systems in an autonomous world can be driven by good old fashioned unit tests. No need to get fancy, just generate hundreds of tests.