Is there anyone here who regularly use LLMs effectively for writing Nextflow pipeline ?

I have had really bad luck asking LLMs any questions about Nextflow. Yesterday one told me

combine

doesn't have a

by

argument...

Yeah me too

My (limited) experience is that sometimes it points you to a nice feature you didn't know existed, and you can then dig the docs and write the code yourself, if this feature actually exists 😅

I see... basically looking for someone to do a #C01LNCGJBMH at the request of @Ammar Aziz but everyone seems to be avoiding it

I have also tried Seqera AI but didn't find it to be much different in outputs

I also don't have great experience with it, LLMs do help with simple code suggestions though but are definitely not at the point where they can effectively write full Nextflow pipelines

Yeah, same here, it rarely produces working nextflow code or capable of simplifying code snippets without changing what’s actually being done entirely.

What's worth adding is that they "know" Groovy quite well, so I often find inputs on that (doing list manipulation, etc.) can be quite helpful

I had an old pipeline with a groovy function in

<http://main.nf|main.nf>

that included a for loop and no

def

for a variable name. Github Copilot (GPT 4.1) was able to refactor that quickly which was nice. But I haven't found use for anything larger than a function or improved tab complete for documentation.

I also had a funny experience with Copilot where it invented some channel operators.

I am using Seqera AI for very specific Nextflow questions. It's not perfect, but I've found it to be quite useful in multiple cases