This is a retrospective of one of our early products — Parview, an enterprise document-archiving tool we built in the late 2000s. The tool itself is long retired; the lessons aren't. I'm leaving this page up as a little archive, and using it as an excuse to share what the work taught us.
The context
In the mid-2000s, a lot of organisations were belatedly realising that their filing rooms were going to kill them. Decades of paper records — student transcripts, property deeds, court filings, medical records — sat in basements, slowly getting eaten by damp, mice, and retirement. The people who knew where things were kept retiring. The ones replacing them couldn't find anything.
We worked with a few universities and government offices on custom digitisation projects. After the third or fourth one, we realised the software layer was largely the same each time: scanner integration, OCR, encrypted storage, full-text search, access controls. So we packaged it up as Parview.
What we learned
Three things, mostly.
Scanning is the easy part. The first project we did, we scoped the scanning as 80% of the work. It was closer to 20%. The actual work was metadata — tagging each document with the context that made it findable later. You can OCR a handwritten student record from 1958, but if nobody marks it as "student records," "1958," "mechanical engineering department," it's lost in the pile as effectively as it was before.
The humans who used to find things are your most important users. Every filing room has a librarian — the person who could, uncannily, find any document in minutes using a mental model nobody else could articulate. The temptation is to replace this person with the software. The correct approach is to build the software around their mental model, with them. The ones we built with their librarians were loved and used. The ones we built without were shelved within a year.
Privacy and permissions are the whole game. Digital archives get used by more people than paper archives, because they're easier. That's the whole point. But it means a tiny mistake in access control exposes things that, in the paper world, were protected by the physical friction of getting to the filing room. We learned to build permissions from day one, not bolt them on later.
What Parview did, in brief
It was a web-based archive. You logged in, searched, and got back your documents. Under the hood:
- Encrypted storage. AES-256 on everything at rest, decrypted only with the user's passphrase.
- Encrypted transit. SSL, in the days when that still felt like effort.
- Geographically redundant backups. Three continents. Seemed excessive at the time. Saved us at least once.
- Role-based access control. Who can see what, down to the document level.
- Proper search. Full-text over the OCR'd content plus all captured metadata fields. Ranged queries for dates and numbers. Even colour queries for image-heavy archives.
A small story from a university
My favourite moment from that era was watching a professor at an Indian university, the day we switched on Parview for their records room. She typed in a student's name, hit search, and had a full set of transcripts, letters, and examiner reports in two seconds.
She sat back in her chair. "You know," she said, "there's a question a student asked me six years ago. About whether her late grandfather had once been a lecturer here." She typed the grandfather's name. The system brought up a 1963 appointment letter. The professor began to cry, quietly, in a way that suggested she'd been carrying that unresolved question for a long time.
We weren't expecting that kind of reaction. It was the kind of reminder, early in our career, that the work we do affects people in ways we can't anticipate, if we build it with enough care.
What we do now
Parview is retired — enterprise document management is a mature category now, with good commercial options. We don't build in this space any more.
But the lessons from that work shape how we approach AI projects today. Build with the librarians. Privacy from day one. Don't mistake the technology for the task. These are all true regardless of whether the capability in question is a full-text search index or a large language model.
If you're thinking about any kind of digital transformation — archives, records, unstructured documents, or the modern version of these problems with AI on top — we'd love to help.
