Language selection
Fast switching with recently used pairs surfaced first, because a clinic sees the same languages repeatedly.
Product engineering
Lyfe Languages
Clinical translation where a wrong word has consequences.
A translation platform built specifically for clinical settings, where general-purpose translation is not safe enough and a verified medical dictionary sits between the model and the patient.
- Sector
- Healthcare & Clinical Communication
- Year
- 2023
- Duration
- 8 months
- Team
- 6 people
- Platforms
- iOS · Android · Web
- Status
- In production
3 tiers
Confidence levels shown
Verified, validated, machine
sub-second
Verified term lookup
Fully offline for the core dictionary
2-way
Audio translation
Speech in, speech out, both directions
Reviewed
Every clinical term
Before it enters the verified tier
01 — The problem
Clinicians were using consumer translation apps in consultations because the alternative was a two-hour wait for a human interpreter. Those apps translate fluently and confidently, including when they are wrong.
The failure mode is specific and dangerous: a general model renders a clinical term into a colloquial near-equivalent. Dosage instructions, consent language and symptom descriptions are exactly where an approximate answer stops being acceptable.
02 — What we did
We inverted the usual pipeline. A verified clinical dictionary is consulted first; only language the dictionary does not cover reaches a general model, and anything the model produces in a clinical register is flagged for review rather than shown as settled.
The interface tells the truth about its own confidence. Verified terms, machine-translated phrasing and community-validated entries are visually distinct, so a clinician always knows which part of a sentence carries institutional backing and which does not.
Fluent and wrong is the worst possible output in a consultation room.
Interface
What it looks like in use
Two of the screens that carry the most weight in daily use, rebuilt here from the production design system.
9:41
EN → ESOFFLINE
“Take one tablet twice daily”VERIFIED
“Tome una tableta dos veces al día”●
“with food” · machine outputREVIEW
92% verified termsDict v4.1
Confidence is never hiddenVerified, community-validated and machine output are visually distinct, so a clinician always knows what carries institutional backing.
Review queue · Clinical linguists14 PENDING
“shortness of breath” · es-MX3 VOTES
“blood thinner” · ar2 VOTES
“fasting glucose” · ur5 VOTES
3CONFIDENCE TIERS
0AUDIO RETAINED
The dictionary compoundsReviewed machine output is promoted into the verified tier, so clinical coverage grows with use instead of staying static.
Capabilities
What the system does
Grouped by the job each set of capabilities exists to do, rather than by which team built it.
Consultation flow
01 · What happens in the roomAudio translation
Speech recognition and synthesis in both directions, with the transcript kept visible so nothing is lost to mishearing.
Phrase shortcuts
Common clinical exchanges available as one tap, pre-verified so the highest-frequency phrases never route through a model.
Offline core
The verified dictionary and phrase set work with no connectivity, which matters in basements and rural clinics.
Clinical safety
02 · Where the product is opinionatedVerified medical dictionary
A curated term base reviewed by clinical linguists, consulted before any general translation happens.
Confidence banding
Verified, community-validated and machine output are visually distinct in the transcript, never blended together.
Translation review queue
Machine output in clinical registers is queued for expert review and, once approved, promoted into the verified tier.
Ambiguity prompts
Where a term has multiple clinical senses, the app asks rather than guessing.
Community and governance
03 · How the dictionary growsCommunity validation
Native-speaker clinicians propose and vote on renderings, with contribution history attached to every entry.
Dialect handling
Regional variants held as siblings rather than overwritten, so the right variant reaches the right clinic.
Change history
Every dictionary entry carries its full revision trail and the reviewer who approved it.
Institution scoping
Hospitals can layer local terminology over the shared base without forking it.
Architecture
How it is put together
Layer by layer, with the reason each one exists — because the reason is usually the interesting part.
01Client
Cross-platform app with a bundled offline dictionary and on-device speech where the platform supports it.
- React Native
- SQLite
- On-device ASR
02Term resolution
A cascade that consults verified terms, then validated community entries, then general translation, tagging output with its tier.
- Cascade resolver
- Tier tagging
03Speech pipeline
Streaming recognition and synthesis with clinical vocabulary biasing to improve recognition of drug and condition names.
- Streaming ASR
- TTS
- Vocabulary biasing
04Dictionary service
Versioned term base with proposal, review and promotion workflows plus per-institution overlays.
- PostgreSQL
- Workflow engine
05Platform
Regional deployment with data residency controls and no retention of consultation audio.
- Regional hosting
- Zero audio retention
Technology
Mobile
- React Native
- TypeScript
- SQLite
- Native speech APIs
Backend
- Node.js
- PostgreSQL
- Redis
- Workflow engine
Speech
- Streaming ASR
- Neural TTS
- Vocabulary biasing
Platform
- Regional hosting
- Docker
- Audit logging
Outcome
What changed
Measured against how the operation ran before, not against a benchmark chosen after the fact.
-
Confidence is never hidden. A clinician can always see which part of a translation is institutionally verified.
-
The dictionary compounds. Reviewed machine output is promoted, so coverage grows with use instead of staying static.
-
It works where connectivity does not. The verified core is fully offline.
-
Consultation audio is not retained. The privacy position is architectural rather than a policy statement.
How it ran
-
Weeks 1-4
Clinical discovery
Observed interpreted consultations and catalogued the failure modes of consumer translation in clinical use.
-
Weeks 5-10
Dictionary foundation
Term base schema, review workflow and the initial verified vocabulary.
-
Weeks 11-20
Resolution cascade
Tiered lookup, confidence banding and the ambiguity prompts.
-
Weeks 21-28
Speech and offline
Two-way audio, vocabulary biasing and the offline bundle.
-
Weeks 29-34
Governance
Community validation, institution overlays and change history.
Related work
From the same practice
Recycling & E-Waste
Razor ERP
One connected system for a warehouse that ran on paper.
A mobile-first ERP for warehouse and e-waste operations that replaced a decade of spreadsheets, paper travellers and radio calls with a single connected system that works on the floor..
2024 · 7 months
Investment & Asset Management
Kilwa
Market data, sentiment and research in one investment workspace.
An AI investment-intelligence platform that pulls market data, news sentiment and analyst research into a single workspace, so an investment team can go from a question to a defensible answer without leaving the screen..
2025 · 9 months
HR Tech & Recruitment
The Walt
Where candidates move, and where they stall.
A recruitment analytics platform that turns an applicant tracking system's raw event log into a picture of where a hiring funnel actually leaks, with the bottleneck named rather than left to be inferred..
2024 · 5 months
Something similar?
Running into the same problem Lyfe Languages solved?
Tell us what you run. We will reply within two business days with what we would build for your situation — and, just as usefully, what we would leave out.