FieldNotes: human-in-the-loop coding for qualitative research

The problem

Coding qualitative data by hand is slow and painstaking: researchers read transcript after transcript, tagging passages against a codebook. Pure-AI tools promise to automate it, but they break the one thing research can't give up — a defensible, auditable trail showing a human made every coding decision.

What we built

FieldNotes applies the researcher's own codebook across their transcripts and proposes codes — but every proposal is approved, merged, or rejected by the researcher. It tracks saturation as coding progresses, keeps a full audit trail of who decided what, and exports the finished analysis in one click. The speed of automation, without surrendering methodological rigour.

Where AI earns its place

The AI does the first pass — surfacing candidate codes far faster than a human scanning line by line — while the researcher stays the decision-maker on every code. That human-in-the-loop design is the whole point: it makes the tool fast and citable in serious research.

Approach & stack

We designed and built FieldNotes end to end — the codebook engine, the approve/merge/reject workflow, saturation tracking, and export — on a modern TypeScript foundation with GDPR-compliant EU hosting, a non-negotiable for research data. The workflow was designed around the researcher's judgement first, with the AI slotted in to accelerate it — never to replace the decisions that make findings defensible.

Outcome

FieldNotes is live and available to research teams. Coding that once meant reading transcript after transcript by hand now moves at the speed of an AI first pass — while the researcher stays the decision-maker on every code, with a full audit trail, saturation tracking, and one-click export. Fast enough to save real time, rigorous enough to defend.