About Steno
Steno started as an internal tool for a historical research group in Hong Kong digitising gazetteers, yearbooks, and directories. Off-the-shelf OCR handled the easy pages and left the rest, and there was no way to turn the corrections historians were already making into a better model.
So we built the loop: annotate, review to a gold standard, snapshot the dataset, fine-tune, serve. Steno is that tool, run as a hosted service for archives, libraries, and public bodies with the same problem.
We are a small team based in Hong Kong. We stay close to the people who use the product and we ship what they need next.
Why Steno
Long before software could read a page, stenographers, most of them women, turned handwriting and dictation into clean typescript and kept offices, courts, and newsrooms running. Their shorthand was the first information technology many workplaces ever had.
Steno takes its name and its mark from that work. The machine reads first; people with domain knowledge make the record trustworthy, and their corrections carry forward.
- Based in
- Hong Kong
- Hosting
- Hong Kong
- Languages
- Chinese (vertical and horizontal), English, mixed-script pages
- Model
- Self-hosted vision-language OCR, fine-tuned per team
No perfect OCR, only fast correction
Historical pages are noisy. We optimise for how quickly an expert can fix a page, not for a benchmark number on clean scans.
Your data trains your model
Gold pages are the asset. They stay with your team and only ever train models your team serves.
Leave any time
Everything you produce exports in open formats. A service you cannot leave is not one you should trust with an archive.