How I built AskRoby

AskRoby exists because for 14 years I ran a creative district with 53 tenants, and the answer to almost every question was inside a document nobody wanted to open.

This is how it was built, what turned out to be hard, and the decision that mattered most.

The problem was retrieval, not storage

Every business I have worked in already stores its documents. Drives, folders, email attachments, a filing cabinet. Storage was never the problem.

The problem is that retrieving one fact costs a person twenty minutes: remember which document, find it, open it, read until the sentence appears. Multiply by a team, by every week, by a year.

So the product was never "somewhere to put documents". It was "ask a question, get the answer".

Reading a document is three separate problems

The first surprise was that "read the document" is not one task. It is three, and each fails differently.

• Getting text out of the file at all. A clean digital contract is easy. A photograph of a signed page taken at an angle in bad light is not, and that is what real businesses actually have.
• Understanding the structure. A 40-page lease is not 40 pages of equal importance. Dates, amounts, parties and obligations matter; boilerplate does not.
• Answering the question that was asked, which is rarely the question the document was written to answer.

The hardest documents were never the complicated ones. They were the badly photographed ones.

The decision that shaped everything: show the page

Early on, the system gave answers. Confident, well-written answers. And I did not trust them — because I knew what happens when a contract question is answered from memory, and this was a machine doing exactly that at speed.

So every answer carries the page it came from. You read the answer, you glance at the source, and either it says what the system claims or it does not.

This changed the product from something you have to believe into something you can check. That distinction is the difference between a demo and a tool somebody uses on a Monday morning with money at stake.

What is actually inside it

• Document upload and processing, including scans and photographs
• Text recognition, so a photographed page is as searchable as a typed one
• AI reading that extracts and structures what is inside each document
• Question answering that cites its source page
• Accounts, two-factor login and workspaces
• Subscriptions and payments
• Expiry alerts for documents that go stale — insurance, licences, contracts
• A full bilingual interface, English and Spanish

What I would tell someone starting the same thing

Build the checking before you build the answering. An unverifiable answer is worse than no answer, because it is confidently wrong and nobody catches it until it costs something.

Test with your worst documents on day one. Clean test data hides every problem you will actually have.

And be honest about when this is not worth it. If you have twenty documents and read them twice a year, you do not need a system. You need a folder and a calendar reminder.