Telling Technology · Learning in public
All episodes
EP.124 — DELETED WHEN DONE
One evening, one product

My transcription service forgets your audio the moment it is done. The code makes sure.

I built a paid transcription product with Claude Code in one evening. Before any feature, it got its retention rules: audio deleted when the job finishes, the transcript on your first download, a 24-hour hard cap on everything else. A privacy page can promise that. Here, a test fails if it is not true.

An hourglass standing on a glowing teal audio waveform on a dark desk; the top half of the glass is already empty, and the sand that ran through has turned into a few lines of text that fade out at the bottom
⌛ 20 commits · 20:50–23:52 · 130 tests · 0 audio files kept
why the rules came before the features
01 The promise

Write down what gets deleted, before writing anything else

People who handle client audio, interviews, meetings, coaching calls, all have the same quiet worry about transcription services: where does my recording go afterwards? So the first thing this product got was not a feature, it was four rules. Audio is deleted the moment transcription finishes or fails. The transcript is deleted on your first download. Anything left over is gone after 24 hours. History keeps metadata only: a file name, a length, a date. Every later piece of code had to fit those four lines.

claude code — desktop
> Before we build anything: write the retention rules for this product. What do we store, for how long, and what deletes it?
Four rules, each with the code path that enforces it: audio gone on done or fail, transcript gone on first download, 24 h hard cap, history is metadata only. Transcript text never lands in the database, the logs, the mails or the reports. (paraphrased)
A table titled what we keep and for how long: your audio file, deleted the moment transcription finishes or fails; the transcript, deleted on your first download or 24 hours after it is ready; your history, file name, length and date only, no content
4 rules · before feature 1
The retention table, as the privacy page shows it.
02 The shape

Audio never touches a cloud AI

The website is a Cloudflare Worker with a small database and a storage bucket in the EU. You upload in 50 MB parts and get a place in the queue. The actual transcription happens on a Mac mini at home, which asks the queue for work instead of being reachable from the internet. It runs Whisper locally: a Swedish-tuned model for Swedish, large-v3-turbo for the rest. A 9-minute file took 61 seconds, about 9x realtime. Every file is capped at 4 hours, and ffmpeg is only allowed to read the file itself, so a booby-trapped upload cannot make it fetch anything from the web.

claude code — desktop
> Someone uploads a playlist file that points at a URL. What does the mini do?
Nothing it should not. Every ffmpeg call carries -protocol_whitelist file,pipe. Tested with a canary URL: the job failed as unreadable and the canary never saw a request. (paraphrased)
A diagram: a browser uploads audio in parts to a Cloudflare Worker with a database and an EU storage bucket; a Mac mini on the right polls the queue, pulls the audio, transcribes it with local Whisper and sends back the transcript, then the audio icon is crossed out
9.2x realtime · 4 h cap
The mini polls. Nothing on the internet can call it.
03 The report

The transcript never waits for the extras

Next to the transcript you get a small report about how it was said: pace over time, pauses over two seconds, filler rate. And a theme sketch, a handful of short labels from a local language model on the same Mac mini. The labels have guards: no names from the recording, no copied three-word phrases, at most four words each. That model shares the mini’s GPU with other jobs and once took 255 seconds. So the order flipped: the transcript ships first, and the themes are merged into the report later. Nobody waits for a nice-to-have.

claude code — desktop
> The theme sketch took four minutes on that last run. The customer is just sitting there?
Not anymore. Transcript first, themes later through a report endpoint that accepts theme fields only. Two synthetic meetings with invented names: 8 labels each, zero names, zero copied phrases. (paraphrased)
A report modal in dark mode: a teal pace line chart across ten slices of the recording, tiles for pauses and filler rate, and a row of theme chips with short neutral labels
transcript first · themes later
Structure report plus theme sketch, never the text.
⚠️

Not live yet, on purpose

Everything in this episode is built and tested, nothing is deployed. The privacy notice, terms and data processing agreement were written from the code, and a person still has to read them before launch. The go-live is a ten-step script that asks before every step, and it is mine to run. Until then the product keeps its working name.

04 The boring parts

What turns a demo into a product

The transcription itself was a few hours of work. Most of the commits went into the parts nobody films: sign-in by magic link, API tokens, a cleanup job that also hunts for orphaned files, billing, account deletion and data export, legal pages that describe what the code really does. And one door built for AI: a local connector so Claude Code can send a recording and read the transcript itself. Tap any image to enlarge it and read the exact prompt that drew it.

BUILT 2026-10-10 · GO-LIVE IS STILL MINE

The promise is a test, not a paragraph

Most privacy promises live on a page nobody reads, written by someone who never saw the code. Here it runs the other way. The rules came first, the code was built to keep them, and the legal pages were written afterwards from what the code does. When I tried to download a transcript a second time, the server said it was gone. That is the whole product in one response.

A bold typographic card on near-black: second download, then a big teal 410 Gone. Underneath: audio deleted on done, transcript deleted on download
20 commits · 93 Worker + 31 poller + 6 connector tests · local Whisper on a Mac mini · EU storage · not deployed yet
06 Steal this

The retention pattern: delete on done, delete on download

The product repo is private while it is a paid service. What I plan to share is the pattern: the retention rules, the purge job with its orphan sweep, the poller loop and the go-live runbook layout. Not published under tellingtechnology yet, so the buttons point at the org for now. Everything in this episode is free and open — clone it, run it, make it yours.

skill the four retention rules cmd purge cron + orphan sweep template Mac mini poller loop assets ten-step go-live runbook
run it # pattern soon. the rule: every code path that ends a job deletes the audio, with a test

No GitHub? Comment FORGET on the post and the bot DMs you the link.

Next episode

The first job moves to the Mac mini, and the keychain says no

Next: one schedule file now decides which machine runs every job. The first pilot moved to the Mac mini, and then macOS refused to hand it the keys.

A darkened schedule manifest where one job row is lit on the Mac mini side, with a small padlock icon next to it
drops soon · follow so you don’t miss it