From handwritten mailbag to drafted reply — AI triage for a magazine’s reader correspondence

A print magazine for an audience that still writes by hand receives a steady stream of letters from readers — questions, advice requests, life updates, deeply personal stories. The editorial team wanted to honour every one with a real reply, but the volume made it impossible to read every letter, let alone respond. Letters piled up; the best ones sometimes never got to an editor at all.

What we built

An AI pipeline that turns the mailbag into structured editorial work. Every letter is scanned, digitised through AI handwriting recognition, and stored alongside its original. The system triages each letter by topic, urgency, and tone, extracts the structured data editors need (reader name, location, age range, life stage, recurring themes), and drafts a personalised reply grounded in the letter’s actual content. Editors review, edit, and approve — keeping the human voice — but every letter gets attention, and no letter gets lost.

How we built it

A Next.js editorial dashboard on Vercel, Supabase as the source of truth for letters, readers, themes, and reply threads. OpenAI handles the handwriting recognition, the topic triage, the extraction, and the reply drafting — with explicit guardrails so the model writes in the magazine’s voice and never invents facts about the letter. Trigger.dev orchestrates the durable jobs. MinIO stores the scanned originals so editors can always go back to the letter as written.

The outcome

A mailbag that used to overwhelm the editorial team is now a structured stream of editorial work. Every reader who writes in gets read; most get a personalised reply within days. The relationship the magazine has with its readers — the thing the magazine is actually about — got bigger, not smaller, as the volume grew.