# AI Translator for Healthcare — Competitive & UX Research **Product:** Real-time, medically accurate translation for clinical workflows **Date:** 2026-05-15 **Purpose:** Ground the UI flow design in real competitor patterns and healthcare-specific UX constraints before sketching screens. --- ## 0. Research Method & Coverage Gaps | Source | Status | |---|---| | Public web (vendor sites, app stores, trade press) | Covered — 8 competitor scans | | Mobbin medical category (`mobbin.com/explore/mobile/app-categories/medical`) | **Blocked** — HTTP 403, paywalled | | Mobbin MCP server | **Not connected** in this session. Mobbin launched an MCP in May 2026 exposing ~621k screens / 142k flows; connecting it would surface real bedside / intake / phrasebook screens. Recommend connecting before high-fidelity design. | | Direct hospital observation / clinician interview | Out of scope for this pass — call out as next step if going to hi-fi | **Implication:** This document covers feature/positioning intel well, and structural UX patterns reasonably well, but is light on pixel-level screen references. The flow we design should be tested against Mobbin once the MCP is available. --- ## 1. Competitor Landscape Five archetypes are competing for the same hospital budget. The product brief points at archetype E (AI-first), which barely exists yet — most market leaders are still human-interpreter-routing platforms. | # | Product | Archetype | Languages | Connect time | Modality | EHR | Notes | | --- | --------------------------- | ------------------------ | -------------------------------- | --------------- | ------------------------- | -------------------------- | -------------------------------------------------------------------------------------------------------------- | | 1 | **Canopy Speak** | A. Verified phrasebook | Spanish, Mandarin (+more in dev) | Instant (local) | Audio + text | – | 5,000 human-QA'd phrases, learning mode, iOS only, by specialty | | 2 | **Care to Translate** | A. Phrasebook + Live | 48 verified / 130 live | Instant | Audio + text + images | – | 3,500 verified phrases, illustrative images, offline mode, low-literacy aids | | 3 | **Martti (UpHealth)** | B. VRI/OPI human | 250+ (90+ video) | Seconds | Video + phone | Yes | IV-cart device w/ 12" tablet, "point to your country/flag" UI | | 4 | **Jeenie** | B. VRI/OPI human | 300+ | <13s avg | Video + phone | Zoom, EMR | Avg 4y interpreter experience, 4-star min, Zoom-native | | 5 | **Voyce** | B. VRI/OPI human | 250+ (60+ video) | <20s | Video + phone | Epic, Cerner, AmWell, Zoom | Auto-documents session in EHR | | 6 | **LanguageLine** | B. VRI/OPI human | 240+ | Seconds | Video + phone | Epic | One-click from chart, language pre-populated | | 7 | **DeepL Voice** | C. AI voice (horizontal) | 40+ | Real-time | Voice-to-voice + captions | Teams, Zoom | "No permanent storage after call", face-to-face mode, healthcare positioned for **intake only** (not clinical) | | 8 | **Google Interpreter Mode** | D. AI consumer | 44 | Real-time | Voice + text | – | Voice-activated, dual-language transcript, Smart Replies | | – | **Our product** | E. AI clinical-grade | TBD | Real-time | Voice + text | Goal: EHR | **White space:** AI-first + medically validated + HIPAA + clinical workflow | ### Strategic takeaway The market is bimodal: **slow but trusted humans** (Martti, Jeenie, Voyce) versus **fast but clinically unvalidated AI** (DeepL, Google). DeepL explicitly *avoids* recommending its product for clinical conversation — only intake. **The defensible wedge is AI-first translation that is clinically validated** (Clinical Clarity Filter), HIPAA-architected, with a human-interpreter escalation path when confidence is low. --- ## 2. UX Patterns Worth Stealing ### From verified-phrase apps (Canopy, Care to Translate) - **Specialty-tagged phrase lists** ("Emergency Medicine", "Pediatrics", "General Surgery") — surfaces the right starter content for the encounter type. - **Verified-translation badge** next to phrases that have been human-QA'd; falls back to "machine translation" label when not verified. **Strong trust signal — directly reusable for our Clinical Clarity Filter.** - **Illustrative images** for procedure/anatomy phrases — helps low-literacy patients and non-verbal communication. - **User-built playlists** for repeated workflows (e.g., a triage nurse's top 20 phrases). ### From VRI human-interpreter apps (Martti, Jeenie, Voyce) - **"Point to your country/flag"** as the patient-facing language selector — works even if the patient can't read or speak any of the offered languages. - **Connect-time copy** ("connected in 8 seconds") is surfaced as the headline number on every marketing page — speed is the #1 purchased benefit. Our UI should show **time-to-first-translation** prominently. - **Auto-populate patient language from EHR** so clinicians never have to pick. (LanguageLine, Voyce.) - **One-click launch from inside the patient chart** — no app switch. Encounter ID, department, provider ID auto-pass to the session. - **Auto-document the session back to the EHR** under a dedicated "Interpreter Information" note — interpreter ID, duration, language. We should do the same: write a structured note + the translated transcript. ### From AI horizontal translators (DeepL Voice, Google Interpreter) - **Face-to-face split view**: phone laid flat between two people, each side of the screen oriented toward one speaker, captions inverted on top half. Critical for bedside. - **Auto-language-detection** on the patient side — the clinician picks their language once; the AI figures out the patient's language from the first utterance. - **Smart Replies** — after the patient speaks, suggested short responses ("Yes / No / I don't understand / Please repeat") so a non-fluent clinician can answer instantly. - **Live captions** in both languages, not just the translation — clinicians need to see *what they said* to catch errors. - **"No data retained after session"** stated visibly. DeepL leads with this for enterprise sales. ### From healthcare design systems generally - **Plain-language consent** in the patient's preferred language *before* the conversation starts — never legalese. - **HIPAA / encryption indicators** as persistent UI chrome (lock icon, "Encrypted session" label) — every screen, not just the splash. - **Session timeout + biometric unlock** to protect any cached transcripts on shared devices. - **Role-based view**: a clinician sees the chart; a registration clerk doesn't. - **Audit trail surfaces in UI** (not just backend) — "Session logged at 14:32, ID #..." builds compliance confidence. --- ## 3. Information-Hierarchy Principles for This Product Ranked by what must dominate the screen in the live-conversation moment: 1. **The translated text the patient just heard / will hear** — biggest type, highest contrast. 2. **Mic state** (who is speaking, listening, paused) — colour + animation, glanceable from a meter away. 3. **Original speech** as smaller caption beneath translation — clinician can audit accuracy. 4. **Clinical Clarity flags** — inline warnings on ambiguous terms ("'discharge' has 3 possible meanings; tap to disambiguate"). 5. **Language pair** — small, but always visible, with one-tap swap. 6. **Compliance chrome** — lock + "encrypted, not stored" — present but not loud. 7. **Escalate to human interpreter** — visible, not buried; one tap. 8. **Session controls** (pause, end, save to chart) — bottom bar, thumb-reachable. The patient-facing half should be **calmer, larger type, fewer controls** — they are anxious and may not be tech-literate. The clinician-facing half can be denser. --- ## 4. Compliance & Trust Signals (HIPAA-Ready) What the UI must show to earn enterprise trust: | Signal | Where it lives in the UI | |---|---| | End-to-end encryption (AES-256 + TLS 1.2) | Persistent lock icon + tooltip on every session screen | | No PHI used to train models | One-line statement on session start + in settings | | Audit log (who, what, when, from where) | "View audit trail" link in session history | | Role-based access | Different home screens per role; never show a chart to non-clinical staff | | Session timeout + biometric unlock | Auto-lock after N minutes idle; FaceID/TouchID re-entry | | Consent capture in patient's language | First screen of every encounter, in patient's language, audio playback option | | Retention controls | Admin setting: "Delete transcript after X" — surfaced in session footer | | Tamper-resistant transcript | Hash + timestamp on exported PDF | **Worth stating outright in product copy:** "Transcripts are not retained after the session unless you save to chart." This single sentence is the most common trust-builder across DeepL, Canopy, and Care to Translate. --- ## 5. Realistic Flow Opportunities (for the actual design step) Ranked by clinical value × design defensibility: ### A. **Bedside conversation** (recommend as primary flow to design) - Highest-volume use case, biggest pain point (waiting for human interpreter mid-rounds). - Face-to-face split view; one device between clinician and patient. - Should show: AI translation + Clinical Clarity flags + "escalate to human" button + save-to-chart on end. ### B. **Triage / intake** - Patient self-serves on a kiosk or tablet before being seen. - Language auto-detect from first utterance. - Structured Q&A — collects symptoms, medications, allergies into a pre-formatted intake note. - Hands off to the triage nurse with the transcript already in the chart. ### C. **Telehealth 3-way** - Provider + patient + AI translator overlay on top of Zoom / Teams / Epic telehealth. - Live captions in both languages on top of the video feed. - Less novel — DeepL/Voyce already in this space — but table-stakes. ### D. **Discharge / medication reconciliation** - High-stakes, structured. Templated discharge script with verified translations for each instruction. - Patient signs back: AI asks them to repeat the instruction in their language, checks comprehension, flags gaps. - Strong outcome story (reduces readmissions tied to misunderstood instructions). ### E. **Review & export** - Post-encounter: clinician reviews transcript, flags any segment for clinical accuracy, exports to chart or PDF. - Audit-friendly — every edit is logged. ### Recommended scope for the design submission **Pick A (bedside) as the hero flow**, with B (intake) and E (review/export) as supporting screens. That trio covers: speed (B), live use (A), and compliance/audit (E) — the three benefits the brief calls out. --- ## 6. Key Assumptions for the Design Stated explicitly so reviewers can challenge them: 1. **Single device, not two.** One phone or tablet between clinician and patient — not a paired-device model. Cheaper to deploy, matches Martti/DeepL precedent. 2. **Clinician is the primary user.** Patient is an important secondary user but doesn't authenticate; they interact through a "patient mode" the clinician hands them. 3. **AI-first with human fallback.** Default to AI; one-tap escalation to a human interpreter when confidence is low or the clinician requests it. 4. **EHR integration is post-MVP** for the design submission, but the flow must show what data would flow to the EHR (auto-documented note, encounter ID, language, duration). 5. **"Clinical Clarity Filter"** = real-time confidence scoring on medical terms, with inline disambiguation when score is below threshold. (E.g., flagging "discharge" → ["leave hospital", "bodily fluid", "electrical"].) 6. **Languages at launch:** Spanish, Mandarin, Vietnamese, Arabic, Haitian Creole, Russian — the six highest-LEP-population languages in US healthcare. Auto-detect across all, manual override available. 7. **Compliance posture:** HIPAA-ready (BAA-able), SOC 2 Type II target, on-device processing where feasible. --- ## 7. Next Steps Before Designing 1. **Connect the Mobbin MCP** and pull 10–20 real bedside / intake / telehealth screens for reference. Specifically search: medical phrasebook, video interpretation, telehealth captions, consent screens. 2. **Decide hero flow.** Recommend bedside (A). Confirm with stakeholder. 3. **Sketch wireframes** for the 5–7 key screens (Splash + consent → Encounter setup → Live conversation → Clarity-flag disambiguation → Escalate to human → Session end / save to chart → Review & export). 4. **Validate with one clinician** before hi-fi. --- ## Sources ### Direct competitors - [CanopySpeak — Canopy Innovations](https://info.withcanopy.com/canopy-speak) - [Canopy: a medical translation app — ENT & Audiology News](https://www.entandaudiologynews.com/reviews/tech-reviews/post/canopy-a-medical-translation-app) - [Martti by UpHealth](https://www.martti.io/) - [Care in Any Language — RWJBH](https://www.rwjbh.org/blog/2023/march/care-in-any-language/) - [Martti device platform](https://www.martti.io/platform/devices) - [Jeenie healthcare](https://jeenie.com/industries/healthcare) - [Jeenie medical interpreter services](https://jeenie.com/medical-interpreter-services/) - [Voyce medical interpretation](https://web.voyceglobal.com/) - [Voyce technology](https://web.voyceglobal.com/technology) - [Care to Translate](https://www.caretotranslate.com/) - [Care to Translate — phrase library guide](https://www.caretotranslate.com/news/medical-phrase-library-a-guide-to-using-verified-translations-in-health-care) - [Care to Translate — GDPR/HIPAA compliance](https://www.caretotranslate.com/news/data-security-in-medical-translation-apps-gdpr-and-hipaa-compliance) ### AI horizontal translators - [DeepL Voice for Conversations](https://www.deepl.com/en/products/voice/deepl-voice-for-conversations) - [DeepL Voice for Healthcare](https://www.deepl.com/en/industries/healthcare) - [DeepL unveils real-time spoken translation](https://www.deepl.com/en/press-release/deepl-unveils-real-time-spoken-translation-breaking-the-next-language-barrier-with-voice-to-voice) - [Google Interpreter mode](https://assistant.google.com/interpreter-mode/) - [Google Blog — Interpreter mode](https://blog.google/products/assistant/interpreter-mode-brings-real-time-translation-your-phone/) ### EHR integration patterns - [Boostlingo — EHR integration benefits](https://boostlingo.com/blog/interpreting-platform-ehr-integration/) - [LanguageLine — EHR interpreting integration](https://www.languageline.com/interpreting-services/ehr-interpreting-integration) - [Martti — EHR & telehealth integrations](https://www.martti.io/platform/integrations) - [JAMIA Open — System-wide digital medical interpretation framework](https://academic.oup.com/jamiaopen/article/8/6/ooaf100/8327119) - [LanguageLine + Epic announcement](https://www.languageline.com/blog/languageline-announces-integration-with-epic-ehr) ### HIPAA / trust UX - [HIPAA-Compliant Translation App — Accountable](https://www.accountablehq.com/post/hipaa-compliant-translation-app-for-secure-patient-communication) - [HIPAA-Compliant UI/UX: 7 Principles — Orbix Studio](https://medium.com/@orbix.studiollc/hipaa-compliant-ui-ux-7-design-principles-for-healthcare-f62796899002) - [Healthcare UX Design Best Practices 2026 — Fuselab](https://fuselabcreative.com/healthcare-ux-design-best-practices-guide/) - [Healthcare App UI Design That Patients Trust — Artonest](https://medium.com/@artonest.design/healthcare-app-ui-design-that-patients-trust-c4693dcccfbf) ### Mobbin (gap to fill) - [Mobbin home](https://mobbin.com/) - [Mobbin medical category](https://mobbin.com/explore/mobile/app-categories/medical) (paywalled — 403 to public fetcher) - [Mobbin Launches MCP Server — Business Wire](https://www.businesswire.com/news/home/20260511053592/en/Mobbin-Launches-MCP-Server-Giving-AI-Tools-621500-Real-App-Screens-to-Reference) - [Mobbin MCP unlocked walkthrough — DesignerUp](https://designerup.co/blog/mobbin-mcp-to-ai-1000s-of-real-ui-screen-designs/amp/)