AI Dent.
4 AI products for dental. From MVP to scale. The AI pipeline worked — doctors couldn’t use it.
Head of Design for 3 years. Under NDA.
read by the AI
- What
- 4 AI products for dental: scan reading, treatment planning, a desktop MIS app and Scribe.
- Why
- The AI pipeline worked, but doctors couldn’t use it: 70% misread the AI tools in first-time tests.
- My role
- Head of Design, 3 years: research, UX/UI, design system, team of 2.
- Result
- 1000+ clinics · 3× faster tasks · −70% “can’t find” tickets
Experts trust AI when they can check it, correct it and act on it.
- Check. Raw confidence scores doctors couldn’t act on became colour-coded overlays, severity tiers and per-tooth finding cards. Before, 70% of newcomers misread the AI tools.
- Correct. Every finding is approved, edited or rejected inline, with a radiologist comment, without leaving the report.
- Act, in production. Used in 1000+ clinics: 3× faster task completion and −70% “can’t find” tickets after the redesign.
An AI that reads dental scans, in an interface doctors actually use — grown from one MVP into four products.
Read the full story
1000+ clinics. Canada, USA, Europe. The AI pipeline worked — but doctors couldn't use it. As Head of Design for 3 years, I rebuilt the entire experience: redesigned UX, built the design system, shipped two themes, and grew the platform from a single MVP to a four-product dental AI suite.
Under NDA. Product identity and client name are confidential. Screens shown are from the published case study on stepikin.ru. Real patient data and proprietary AI model internals are not disclosed. The platform is referenced here as “AI Dent.”
Not one product. Four.
Each with its own UX and research.
AI Radiology & Patient Management
CBCT, panoramic and bitewing analysis; patients, reports, multi-clinic.
Studio — Treatment Planning AI
Prosthetic and orthodontic planning from scan data.
Desktop App — MIS Integration
Native app bridging the AI with clinic management systems.
Scribe — AI Appointment Recording
Records consultations, extracts findings, fills patient notes.
Read the full story
Over 3 years the platform grew from a single AI radiology tool into a full dental AI suite. Each product required its own UX approach, user research, and design system work.
- AI Radiology & Patient Management — CBCT, panoramic, bitewing AI analysis with findings overlay. Patient records, report ordering, multi-clinic management. The main focus of this case.
- Studio — Treatment Planning AI — AI-assisted prosthetic and orthodontic planning. Visualizes restoration scenarios and implant placement from scan data.
- Desktop App — MIS Integration — native desktop app connecting AI Dent to clinic management systems (МИС). Bridges AI analysis with existing clinic software.
- Scribe — AI Appointment Recording — AI recording and transcription of dental consultations. Extracts key findings, auto-populates patient notes.
Built for the AI. Not for the doctor.
Launched in 2022 to validate the AI. Growth turned six UX problems into blockers.
corridor tests
Overloaded patient lists
4–7 clicks to find a patient.
Bloated patient pages
Up to 11 tabs; findings buried.
Unintuitive AI tools
Misunderstood by 70% of newcomers.
Hard 2D / 3D switching
Radiologists lost context mid-report.
Dark only, no mobile
Broke below 1280px.
No design system
Every sprint, new patterns.
Read the full story
The platform launched in 2022 validating the AI pipeline as its north star. As it expanded internationally, six UX problems compounded into critical blockers.
- Cognitive overload in patient lists. Dense, unfiltered tables with no hierarchy. Finding a patient took 4–7 clicks. Doctors spent more time navigating than diagnosing.
- Bloated patient pages. Up to 11 tabs per patient with inconsistent IA. Critical AI findings were buried. Doctors routinely skipped entire sections.
- Unintuitive AI tools. Projection tools, tooth formula widgets, 2D/3D navigation — misunderstood by 70% of first-time users in corridor testing.
- Complex 2D / 3D switching. Moving between panoramic, bitewing, CBCT, and 3D views required multi-step flows. Radiologists lost context mid-report.
- Dark mode only — no mobile. No light theme. Interface broke below 1280px. Unusable on calibrated white exam monitors and in bright clinic lighting.
- No design system. Each sprint introduced new visual patterns. No component library, no token system. Inconsistency compounded with every release.
20+ dentists. Every specialty.
Observation in working clinics on three continents, across four specialties.
“I open the report, but I never know which part matters. Everything looks equally important.”
“I found my patient in two clicks. Before, it was 30 seconds every single time.”
Read the full story
Discovery sessions. In-depth, one-hour observation sessions with radiologists, therapists, prosthodontists, and orthodontists — in their actual clinic environments across three continents. We tracked how doctors actually moved between tools — not what they reported in surveys. Amplitude analytics tracked behavioral patterns at scale post-launch. 20+ interviews across 4 dental specialties.
Continuous validation. A/B interface tests, 5-second comprehension tests for navigation, corridor usability studies. Hypothesis testing before every major release. Post-launch feedback interviews at the 3-month mark.
“I open the report, but I never know which part matters. Everything looks equally important.” — Radiologist, 8 years clinical experience.
“The new patient list — I found my patient in two clicks. Before, it was 30 seconds every single time.” — Prosthodontist, Moscow · post-launch interview.
Research → analytics → design → ship.
Three parallel workstreams at all times. No stage-gated waterfall.
In-clinic discovery
Observation, corridor tests, post-launch interviews.
Amplitude tracking
Funnels, baselines, support tickets as a confusion proxy.
Figma → production
Token-based system, prototypes, annotated specs.
The three workstreams in detail
Qualitative research — in-clinic discovery.
- Observation sessions across 4 dental specialties
- Corridor usability testing with real patient flow
- A/B layout and comprehension tests
- Post-launch interviews 3 months after each major release
Quantitative analytics — Amplitude tracking.
- Event-level behavioral tracking across all key flows
- Funnel analysis for report ordering and patient navigation
- Hypothesis testing against pre-launch baselines
- Support ticket volume as confusion proxy before/after
Design & prototyping — Figma → production.
- 200+ component design system with token-based theming
- Interactive prototypes via Figma Make for usability testing
- Design spec delivery with developer annotations
- Design leadership: reviews, QA, cross-product consistency
Steps: Discover (clinic observation, 20+ interviews, Amplitude analysis) → Define & Map (JTBD, user journeys, information architecture rework) → Design (wireframes → high-fidelity → Figma Make prototypes) → Test & Ship (usability testing, A/B experiments, live tracking).
4–7 clicks to a patient. Now two.
Twelve iterations of layout, filters and density.
- 01Inline searchInstant filter by name, date, status.
- 02Smart sortingLast visit, appointment, report readiness.
- 03Compact / expandedSaved per doctor.
- 04Quick actionsOpen, schedule, share — from the row.
Read the full story
Navigation rebuilt from scratch. Original patient list: 4–7 clicks to reach any record. We ran 12 design iterations — testing layout, filter logic, card density — before landing on a 2-click solution.
- Persistent inline search with instant filter by name, date, status
- Smart sorting: last visit, upcoming appointment, AI report readiness
- Compact / expanded view toggle saved per doctor preference
- Quick-action row — open report, schedule, share without entering patient card
Navigation · IA rework · 12 iterations.
Confidence scores became a diagnosis.
Colour-coded overlays, severity tiers, per-tooth cards — approved in one tap.
Read the full story
AI diagnosis made readable. Raw AI confidence scores doctors couldn't act on → structured diagnostic surface. Color-coded overlays, severity tiers, per-tooth finding cards, one-tap approve — no context switching.
- Panoramic view with AI overlay toggle (show / hide masks)
- Non-dental findings panel — separated from dental pathology
- Per-tooth cards: condition label, AI confidence, radiologist comment
- Inline approve / edit / reject — without leaving the report
AI visualization · Medical UX · Data hierarchy.
11 tabs. One scannable page.
Team, report status and “Create report” first.
Read the full story
Patient card redesigned for clarity: 11 tabs → single scannable page. The new card surfaces what doctors need first: treating team, AI report status, “Create report” action — all without scrolling. Then: report grid, photos, history.
- Treating doctors and shared access visible at a glance
- AI report type grid: CBCT, Pano AI, IOXRay, 3D, Implant, Ortho, Scribe
- Upload flow from patient card — no nested modal chain
- Dental photo gallery with per-image status tags
Information design · Light theme · Workflow.
Five screens. Three steps.
How ordering works
A 5-screen ordering flow with no progress indicators became a single guided modal. Select scan → upload files → confirm. Ordering time: ~4 minutes down to under 60 seconds.
- Step 1: Select CBCT scan from patient history or upload new
- Step 2: Upload optional STL model for 3D reference
- Step 3: Upload dental photos — drag-and-drop with folder support
- Confirmation with estimated AI processing time
Task flow · 3-step modal · Upload UX.
Two themes. One design system.
Dark for radiology monitors, light for bright clinics — from one token library.
Auto-layout, variants, states.
Validated in both themes.
Read the full story
Radiology monitors are calibrated for dark. General dentistry offices run on white walls and bright lighting. Both themes come from a single token-based library — 200+ components, one source of truth.
- Semantic color tokens — one library, two theme outputs
- 200+ Figma components with auto-layout, variants, interactive states
- WCAG AA contrast compliance validated in both themes
- Per-clinic theme preference saved in account settings
Design system · Figma tokens · Accessibility.
Plan the implant. Then talk it through.
Studio — treatment planning AI
Place implants in 3D on CBCT data and preview restorations before any procedure.
Desktop app — inside the clinic’s own software
Periodontal chart and an AI consultation plan, right in the MIS record.
Read the full story
Studio — Treatment Planning AI. Prosthetic and orthodontic planning tool built on CBCT scan data. Doctors place implants in 3D, visualize bone density, and model restoration scenarios — before any procedure starts.
- 3D jaw reconstruction with interactive layer tree (Teeth, Anatomy, Implants)
- Implant manufacturer catalog — select type, size, placement angle in 3D
- Panoramic cross-section synced to 3D model for reference
- Wax-up modeling layer for restoration preview
3D medical UX · Treatment planning · Implant workflow.
Desktop App — MIS Integration. Native Windows app bridging AI Dent into existing clinic management systems. Doctors see AI analysis, periodontal charts, and AI-generated consultation plans — directly inside their existing workflow.
- Periodontal chart with full-mouth tooth visualization + probing marks
- AI-generated consultation plan: per-tooth issues, discussion agenda, next steps
- IOXRay, Pano, CBCT image strip at the bottom — always accessible
- Integrates with MIS patient record — no context switching between apps
Native desktop · MIS integration · Periodontal UX.
Clinics aren’t desktop-only.
Reports between appointments, tablets in consultations, phones on rounds.
Read the full story
Doctors check reports between appointments, present findings on tablets during consultations, and access patient status on phones during rounds. We rebuilt the responsive layer from scratch — both themes, all screen sizes.
Light + dark, every screen size. Both themes built from the same token system — zero separate mobile design pass. The design system adapted automatically from radiology workstations (2560px) down to iPhone SE (375px). Per-clinic theme saved in account settings.
- Responsive from 375px to 2560px — no layout breaks
- Bottom nav replaces sidebar on mobile — thumb-reachable primary actions
- AI report viewer: panoramic → scrollable, 3D → pinch-zoom
- Patient sharing flow optimized for presentation on tablet during consultation
Numbers that matter.
Amplitude, interviews and support tickets, before vs. after.
Clinics in Russia, EU, Canada, USA.
Faster tasks after the navigation redesign.
“Can’t find” tickets, 3 months post-launch.
AI products, 0 to live.
Components shared by all four.
Interviews, 4 specialties.
How it was measured
- 1000+ clinics connected across Russia, EU, Canada, and the United States → international expansion enabled by the full redesign.
- 4 AI products shipped: Radiology, Treatment Planning, Desktop MIS, Scribe AI → designed from 0 to live over 3 years as Head of Design.
- 3× faster task completion in usability testing after the navigation redesign → A/B test · 30 participants · corridor + lab sessions.
- −70% fewer “can't find” support tickets — measured 3 months post-launch → tracked via Amplitude + support ticket logs.
- 200+ Figma components in the shared design system — used across all 4 products → token-based, dark + light, interactive prototypes on Figma Make.
- 20+ in-depth clinical interviews across 4 dental specialties in 3 countries → radiologists, therapists, prosthodontists, orthodontists.
“I found my patient in two clicks.”
Thank you.
Head of Design, 3 years — Alex Stepikin · Stepikin Studio
