Diagnocat
Designing an app for people who don't trust modern design
Deliverabels
Team
1 UX designer
5 software engineers
My roles
UX designer
UX researcher
Tools
Figma
Three.js
TL;DR
Diagnocat is a cloud-based dental imaging platform for clinicians trained on extremely old desktop software that introduces outputs that are editable, shareable, and understandable for patients and referring doctors. AI assistant can analyze radiological reports and detect pathologies and offer treatment pre-planning. Diagnocat allows collaboration between practices and second opinions.
Challenge
GPs and endodontists
Oral surgeons
Pediatric dentists
Periodontists
Prosthodontists
Orthodontists
Labs
Dental industry has one of the highest adoption friction of new software in the world. Most practices are using legacy desktop tools and focusing on time-per-client. The hardest part of shipping a new product in this category is not drawing yet another dental imaging viewer. Earning trust of clinicians stems from rejection of conventional UX best practices and focusing on what's native to their industrial standards, all while keeping context where interpretation is high-stakes and patient data is sensitive.
Solution
My design direction was to move Diagnocat from “AI that detects things” to a clinical workflow system: familiar enough for everyday dentists, powerful enough for specialists to switch from desktop software, time-saving for practices. Making data structured so clinicians could review, correct, measure, annotate, export, share, and integrate findings instead of treating AI output as a black box. 55 leading dentists and oral radiologists informed development, as well as extensive user feedback and real-case testing.
Force of habit
Problem: dentists were used to long-standing native imaging tools, most designed in early 2000-s. They were used to modern mobile apps like WhatsApp or Instagram, but not mobile web-apps. During user-testing stage even basic actions like finding "share my screen" button were difficult to my respondents.
Solution: I conducted deep interviews with general practitioners, prosthodontists, endodontists, orthodontists, and facial surgeons across Israel, US, Europe, and Japan.
Deep interviews
Conducted with dentists and specialists across multiple regions, uncovering differences in workflows and expectations shaped by legacy tools
A/B Testing
Tested key user flows and interface patterns to validate usability improvements and ensure new interactions felt intuitive
Heuristic surveys
Collected feedback through lightweight surveys, helping prioritise improvements and validate design decisions at scale.
Collaboration and second opinions
Problem: dental treatments are rarely done by one person. Cases move between dentists but older workflows make this slow and fragmented. Industry is dependent on “second opinion” and peer-to-peer case sharing is crucial.
Solution: I embedded collaboration into a workflow: clinicians can share a patient file with another doctor, add a comment, invite an external specialist by email, and generate an access code. Collaboration becomes faster and more clinically useful because the shared object is not a raw scan dump, but a reviewed case with findings and comments. That makes Diagnocat feel less like isolated software and more like a clinical handoff system.
AI detection
Problem: CBCT (Cone Beam Computed Tomography) is clinically powerful, but hard to actually see what matters. Manual CBCT review is slow, cognitively heavy, and difficult to standardize. Viewers treat the data differently, segmentation is time-consuming.
Solution. I pushed the viewer toward clarity and teachability. The AI layer on top of the imaging accelerates the workflow without removing clinician control. AI layer does automatic segmentation of teeth, detects 40 different conditions on 2D images and more than 60 on 3D, highlights areas of concern automatically, color-codes tooth states, surfaces suspicious teeth, and lets the clinician accept, correct, or refine findings.
Outcome. The value is not “AI replaces diagnosis.” It is "AI helps clinician with faster screening, more consistent review, and a cleaner handoff into treatment planning and reporting".
Annotation workflows
Problem: dentists do not just review scans; they explain, measure, mark up, and defend decisions. A fixed prediction overlay is not enough for actual clinical work.
Solution: I designed the image workspace as editable. Each tooth card has tools for measurements, arrow markers, comments, links, and additional images. Clinicians can also create their own slices and add them to the report.
Client-readable reports
Problem: reporting is where many dental tools collapse into jargon and clutter. The product has to work both for the clinician and for the patient who needs to understand the next step.
Solution: I simplified the report path into a review-and-finalize flow. One-click report can be signed by a GP and printed. That feature alone contributed to treatment acceptance by up to 23%.
Just-in-time onboarding
Problem: traditional product tours, walkthroughs and onboarding tooltips interrupt the exact task clinicians are trying to finish, which makes them more likely to be dismissed than learned.
Solution: I added a skip-tour option and a folding animation that collapses the onboarding into a persistent help icon, so users can relaunch the tour when they actually need it rather than when the system guesses they might.
Outcome: Discoverability stays available without hijacking clinical flow. For software used under time pressure, that is the right trade-off.
Progressive rollout
Problem: a new feature called dental photo management would have required storage, naming, metadata parsing, foldering, search, filtering, moving, bulk operations, and sharing. Too much for one release.
Solution: I decomposed that scope into seven iterations, beginning with a much simpler grid and progressively adding the more complex operations. From plain to complex operations.
Localized features
Problem: dental notation is not universal. Scans created in the US can have teeth associated with other numbers in quadrant of human jaws than the one made in Europe.
Solution: I proposed an additional ML pass to establish tooth identity, paired with a global setting for the main notation variants.
Ready to build something amazing together?
I'd love to connect with you.