Arbo

Getting a materials quote from timber suppliers meant manually reading building plans, listing every beam and panel by hand, and emailing multiple distributors: slow and error-prone for both builders and Arbo's own team.

The platform has two very different sides. Clients get a simple, guided request flow: upload your plans, and Avi (Arbo's AI) extracts the material list automatically, reviewable before it's sent anywhere. Internally, Arbo's specialists work from a denser operations dashboard: matching requests with consulted suppliers, comparing incoming offers, and managing accounts across three roles (Admin, Expert, Supplier). I designed both sides on a shared design system built with Tailwind CSS and Anima, so the specialist and client screens stayed visually consistent while serving very different needs.

Every request lands in a queue Arbo's specialists work from directly. Before anything moves forward, it goes through a completeness check across four areas, general information, transport, dates, and any special requirements, flagging what's missing before it's processed. Once a request passes, extraction turns it into a structured materials list, broken down by panel type, thickness, surface quality and area, each row carrying its own AI confidence score, so specialists can see at a glance which figures to trust and which to double check before quoting.

Once a request is approved, it becomes an order specialists track through its own pipeline: sent to experts, assigned, sent to suppliers, and through to an offer sent back to the client. Each order carries an automatic summary of its documents, generated by the same AI that extracts the materials list, along with customer and delivery details at a glance, and a record of every supplier consulted, tracked by whether they've opened the request and the status of their quote. Once offers come back, specialists filter and compare them by price and delivery date before sending the best ones on to the client.

Clients get the simpler side of that same flow: upload the files for a request, fill in delivery details while Avi extracts a dimensions-based summary in the background, then review and download the generated list once it's ready. Specialists work from a fuller version of that same extraction internally, the confidence-scored breakdown above, which is what actually drives the quote.

Beyond the shipped flow, I explored a more ambitious redesign: a 3D structure editor where AI would flag reliability issues, like beam collisions, directly on the model before a request was ever sent to a supplier. It didn't make it to production, but it shaped how I think about surfacing AI confidence and errors inside complex technical interfaces.

The AI-assisted request flow replaced a manual, email-based process with a single upload-to-quote flow, reviewable by both the client and Arbo's team before it ever reached a supplier. Automated a process that previously required manually cross-referencing multi-page construction plans, freeing engineers to focus on validation instead of data entry.