Lithai Pletain

RealityScan allows users to create 3D models from photos and video. However, post-scan cleanup — especially cropping and mesh refinement — creates friction and often pushes users to external tools.

This case study focuses on the mobile redesign of the crop and mesh editing tools, aiming to improve precision, clarity, and user confidence during post-scan workflows.

RealityScan serves a wide spectrum of users — from professionals building game assets to hobbyists printing their first 3D model. Understanding who was abandoning the crop tool shaped our priorities.

The existing interface had too much going on and not enough clarity. Users felt they might break their scan, so they left.

Competitors offered basic 3D cropping but lacked integrated mesh editing. Most experiences were fragmented and difficult for non-expert users — an opportunity to go further.

First iteration — Early concepts explored improvements to both mesh editing and cropping interactions, focusing on usability and reducing complexity. The initial crop view still required simultaneous two-finger precision on a small screen, which created more frustration than confidence.

First iteration — crop and mesh tool explorations

What we changed and why:

The 3D gizmo wasn't the first direction we explored. Early concepts tried gesture-based crop adjustments — pinch to expand, swipe to trim — but testing showed users wanted single-axis control, not ambiguous multi-touch on a small screen. A gizmo with one draggable face per axis gave precise control without requiring 3D expertise. Pairing it with undo/redo and a "restore original" option removed the fear of making irreversible mistakes — which was the real reason people were abandoning mid-edit.

The final design simplified interactions and improved precision through clearer controls and better visual feedback. Four screens tell the story:

Final design — crop and mesh editing, iOS

Lighthouse began as an internal operations tool for REEF Technology — a company managing urban parking lots converted into last-mile delivery hubs and pop-up kitchens. As the platform evolved into a standalone SaaS product with an AI co-pilot, it had to serve a much broader and more diverse user base.

The core problem: every new user was funnelled through the same onboarding flow, regardless of who they were or what they needed to do. A real estate partner listing a property was walked through the same steps as an operator who would use the platform every day — a flow designed for neither.

Feedback from user interviews and support tickets pointed to the same friction — the onboarding felt designed for someone else.

Before landing on role-based branching, we explored improving the existing single flow — simplifying the language, reducing steps, adding a progress bar. Testing showed marginal gains. The real problem wasn't the length; it was that the content was wrong for most users. An operator doesn't need to see space submission steps, and a brand partner doesn't care about location scanning. The decision to branch early — one question, three completely separate flows — came from accepting that no single onboarding could serve all three jobs-to-be-done.

The solution was a role-aware, non-linear onboarding system. Instead of one long flow, users are asked a single question at the start — who are you? — and routed into a path built specifically for their context and goals.

Three distinct paths, each with its own scope, content, and completion criteria:

Role picker screen — add lh-role-picker.jpg

The single entry point — one question routes each user into their own path

The real estate path was the most technically novel — it required partners to submit a 3D scan and 360° images of their property using only their phone. We designed a guided scanning experience that walked first-time users through the process with no prior training.

The flow: referral link or QR → role selection → account & business info → scanning walkthrough → scan review & submission.

Key screens from the real estate onboarding path

After launch, qualitative feedback showed a marked shift in sentiment. Users no longer felt lost — they felt the product was built for them.

Lighthouse's AI insights lived entirely on desktop. But the people meant to act on them — kitchen operators, regional managers, repair crews — rarely sat at a desk. They checked their phones in the gaps between tasks, and many weren't comfortable with data tools.

The insight existed. It just wasn't reaching the moment of decision.

By the time an operator got back to a desk, the window to act had usually closed. A location dipping below its 5pm goal, a delivery platform's orders sliding — these were live, shift-level problems. Insights arrived as history, not as something you could still change.

Before designing anything we spent time with the operators. Two things came up in nearly every conversation:

Bringing AI insights to mobile wasn't just a screen-size problem. It raised harder questions:

First approach — standard chat. Our first version was a text box and bubbles. In testing it fell flat: users treated it like a search bar, didn't trust its answers enough to act on them, and the interface had no point of view — so neither did its advice.

Iteration 1 — initial chat UI · users felt it had no point of view

What the feedback told us — three changes:

Second iteration → Final iteration — Athena as named mentor

Before building a persona, we explored improving the chat interface itself — better formatting, faster responses, smarter follow-up suggestions. Users engaged more with the information, but still didn't act on it. The insight was that trust, not clarity, was the gap. Operators were skeptical of faceless AI recommendations mid-shift: if it's wrong, they're the ones accountable. Giving the assistant a name, a consistent voice, and a defined point of view gave users something to hold the AI to — and something to trust. The four principles that followed were guardrails to keep Athena honest.

Named after the Greek goddess of wisdom, Athena embodies knowledge and approachability. She communicates directly yet warmly — built on four guiding principles:

Final design — Athena responds with interpretation, one action, and data you can dig into