Marginalia @ n8n

qualitative research takes time, and deadlines rarely allow for it.

when time runs short you take the route that gets you there, which isn't always the right one. i wanted the repetitive parts of a sprint handled for me.

i split the busywork from the thinking.

before building, i mapped the whole sprint: kickoff, fieldwork, synthesis, delivery. then i sorted the parts that are scaffolding (scheduling, transcripts, a first pass at themes, report structure) from the parts that need a researcher. marginalia handles the first and leaves the second alone.

nothing goes out without me reading it.

every ai step gives a draft, and i question it, because sometimes it simply doesn't make sense. severity scoring is the clearest case: it suggests candidates and i make the call. the workflows are built so the researcher can't be dropped out of the loop by default.

it works however you already work.

run the whole sprint, start from fieldwork, or bring your own transcripts and go straight to synthesis. it isn't tied to one ai provider or transcription tool, so you can swap in whatever your team uses.

this was the hard part.

n8n has a steep learning curve, and my flows kept breaking over small things. it went through several versions, and i kept a fix checklist to work through problems in order. once i understood how the tool thinks, it stopped fighting me.

mostly me, so far.

i use it on my own client work. i also shared it with an interviewer, who thought it could help with the research she does at her company.