Construction QA teams already have inspection processes. They usually live in PDFs, spreadsheets, or exports from other tools.
This work focused on turning those documents into usable inspection templates without manual setup. The goal was simple. Help teams get inspections live faster with less admin and fewer errors.
Visi AI lets users upload an ITP PDF and generate a structured template inside Visibuild. Steps, requirements, and metadata are created automatically. Users review it, make small edits if needed, and publish.
Shaping the problem
I worked closely with a Product Manager to shape the problem before designing a solution.
We mapped how customers and CS teams create templates today. We looked at where time was lost and where errors crept in. We pushed hard on scope to avoid building AI that felt clever but did not help.
We landed on a clear outcome. Translate existing inspection documents into Visibuild templates accurately and quickly.
The problem
Creating inspection templates was slow and repetitive.
Teams started with an ITP PDF. From there, templates were either rebuilt by hand in the UI or passed to CS to be converted through a spreadsheet and admin tools. Both paths took time and introduced risk.
Small formatting differences between documents led to inconsistent templates. Large projects made this worse, with dozens of templates and frequent revisions.
The information already existed. The problem was the process needed to turn it into something usable.
My role
I led product design across the feature. That included working with the PM on problem framing, designing the upload and review flow, shaping how AI output is shown to users, and defining how the system should fail when output quality is poor. I worked closely with engineering through build and iteration. This was end to end product design work.
The approach
The AI was designed to translate, not invent. It takes what already exists and converts it into Visibuild's structure.
Trust mattered more than novelty. Users needed to understand what was generated and feel comfortable editing it before publishing.
Speed mattered too. The main success moment was seeing a usable template appear with minimal effort.
The solution
Users upload an ITP PDF and generate a template in one step.
The output includes a template title, step titles, descriptions, requirements, step types, and relevant metadata. Everything is editable before publishing. Once published, the template is ready to use in inspections straight away.
The system is grounded in real inspection data from across Visibuild. That grounding helps it recognise what a good inspection template looks like in practice.
The impact
Before VisiAI, creating inspection templates meant manually reading long PDFs, interpreting requirements, and rebuilding everything inside Visibuild. The workflow was slow, inconsistent, and heavily dependent on individual interpretation, which often led to rework.
The improved VisiAI flow shifts this from manual setup to assisted creation. Users can now upload a PDF and receive a structured draft template directly in Visibuild. Instead of starting from a blank state, teams begin with a usable foundation that they can review, edit, and approve.
This reduced setup time, lowered cognitive load, and improved consistency across templates, while keeping users in control of the final outcome. AI acts as an accelerator, not a decision-maker.
What I learned
Strong AI features start with clear problem framing.
Working closely with a PM early helped avoid overbuilding. Domain knowledge mattered more than model capability. Assistive AI built trust faster than full automation.
This work also set the foundation for future document based workflows across the product.