AI Document Onboarding

The problem

Onboarding new accounts onto the platform meant manually uploading and categorising hundreds of documents, a slow and error prone process. The onboarding team had no tooling to help and no way to extract structure from existing documents automatically.

My role and the idea

I proposed and led a hackathon project to use AI to read uploaded documents, extract key fields, propose a matching template with a confidence score and save it to the template library, reducing what took hours to minutes.

I conceived, scoped, designed and shipped it independently. The MVP covered a single upload, AI extraction and the template save flow.

What I designed

• Upload: drag and drop with an 8MB limit, a clear progress state and error handling for unsupported file types.
• AI extraction view: the extracted fields shown next to a proposed template match, each with a confidence score so the user can trust or override the suggestion.
• Review and confirm: the user edits any incorrect fields and saves to the template library; the happy path accepts everything in one click.
• Error states: designed for missing fields and low confidence matches, prompting the user to generate a new template rather than failing silently.

Design principle

AI as a first draft, human in control. Confidence scores make the AI's certainty transparent, so users know when to trust it and when to check.

“AI as a first draft, human in control.”

Outcome

• The full MVP shipped within the hackathon session.
• The workflow, prototype and documentation were ready for the onboarding team to act on.
• I presented to stakeholders at the end of the session.
• It positioned me as a driver of AI first design thinking in the UX team.

Reflection

The hackathon reinforced that a small, well scoped MVP built in a day can move a real business problem further than weeks of discussion.