Skip to main content
← All posts

Digits Core Service is the Clinical API Layer

Medical intelligence is separating from the application layer. Corti does it for clinical speech, Autoderm does it for dermatology, Infermedica does it for triage — Digits Core Service does it for the hand. A position paper on why the bundle of intelligence, validation, regulatory wrapper, and access surface is the structural product, not the patient-facing app.

Market positioning

Corti does not run a hospital, does not sell an EHR, and most patients will never know the company exists. It packages medical speech recognition, clinical coding, and ambient documentation into APIs that other healthcare companies embed into their own products. The interface stays with the partner. The intelligence comes from underneath.

Digits Core Service occupies the same layer — for the hand. Hand kinematics, ROM, dexterity, grip, swelling, tremor, and pain become programmable clinical primitives that telemedicine platforms, EHRs, pharmacies, OT/PT networks, and trial sponsors can call.

The shift

Medical intelligence is separating from the application layer

For most of modern healthcare IT, shipping a new clinical capability meant assembling three very different things yourself. You built the model. You secured the regulatory clearance. You built the distribution. Each effort moved on a different clock, required a different team, and consumed a different kind of capital. The fragmentation is most of why healthcare innovation moved slowly for so long.

Those layers are now collapsing into a single product — packaged intelligence, validation, regulatory wrapper, and access layer, delivered together as something another company can directly embed. A pharmacy app no longer needs to become a hand-therapy company to triage hand pain. A hospital platform does not need an internal computer-vision team to ship ROM measurement. An EHR vendor does not need to spend five years training pose-inference models from scratch. They can rent the capability instead, and increasingly they do.

Old model

Build every layer yourself

  • Train and validate the model in-house — multi-year clinical study budgets, full ML org
  • FDA / CE clearance owned by the same team shipping product
  • Enterprise sales motion to hospitals, payors, and EHR vendors
  • Workflow integration rebuilt per customer; long procurement cycles
  • Innovation gated by the slowest of the four — usually regulatory
Clinical API Layer

Intelligence as embeddable infrastructure

  • One company builds, validates, and clears the clinical model
  • Many companies embed it through an API/SDK, keep their UI, workflow, and customer
  • Distribution stays with whoever already owns the workflow
  • Defensibility moves to data depth, workflow integration, and clinician trust
  • Hyperscaler inference becomes the floor — the wrapper is no longer the moat
Where Digits sits

The Digits Core Service is the hand-rehab Clinical API Layer

Digits is not trying to own the patient-facing UI in every market it serves. The web and Android apps are reference experiences — proof that the stack works end-to-end and a direct channel for patients and clinics that want it. But the structural product is Digits Core Service (DCS): a clinically validated, regulator-aware API that turns a phone camera into a hand-assessment instrument and exposes the output as structured medical data.

DCS bundles four things into a single integration: the kinematic intelligence (MediaPipe-driven ROM, dexterity, grip, swelling, tremor, pain scoring), the validation evidence (the Digits Research Connect pipeline that ingests video and produces FHIR-mapped outputs), the regulatory wrapper (clearance posture, audit, consent, PHI controls), and the developer surface (REST + WebSocket APIs, SDKs, Code Connect). One bundle, embedded once, governed once.

1 — Application layer (the partner's product)

Owns the customer
Telemedicine platforms
Hand triage and follow-up in their existing video visit flow
EHR vendors
ROM and dexterity in the chart, written back as Observations
Pharmacy + retail health apps
At-home arthritis tracking inside the app patients already open
OT / PT networks
Pre-visit assessments, between-session homework, outcome reports

2 — Digits Core Service surface

Embeddable
REST APIs
/assessments, /reports, /biomechanics, /lookup, /billing
Realtime WS
/api/ai/ws — voice-guided assessment + live feedback
Web + Android SDKs
Drop-in capture surface; partner keeps brand and chrome
FHIR + verifiable credentials
Observation / DeviceMetric out; consent and audit handled

3 — Digits intelligence core

What gets rented
Hand kinematics
MediaPipe pose + Digits-trained models for ROM, dexterity, claw, fist, abduction
Symptom scoring
Pain v2, arthritis questionnaires, swelling, tremor profiles
Voice clinician
Gemini 2.5-flash orchestrator + MedGemma clinical reasoning
Report generation
AI-summarized PDFs, longitudinal trends, clinician-signable

4 — Data + evidence substrate

Authoritative
Postgres (Prisma)
Per-joint, per-finger validated metrics; ~60 tables
DRC research pipeline
Video ingest → face removal → CSV + interpreted ground truth
Longitudinal cohort
Real clinical deployments compound into validation depth
Outbox to substrates
Optional myLaminin bridge for trial-grade FHIR + ledger anchoring

5 — Regulatory wrapper

Sold once
Clearance posture
Clinical-grade vs. patient-grade output tiers; clearance scoped to the right tier
Auth0 + RBAC
Roles claim, patient/care_provider scopes, JWKS-verified JWTs
Audit trail
Every assessment hashable + e-signable; 21 CFR Part 11-ready via substrate
Consent + PHI controls
Holder-controlled sharing; no public 6-char codes in trial mode
What it looks like in market

The interface is theirs. The intelligence is ours.

The pattern is the same one Corti runs for medical speech, Autoderm runs for dermatology triage, and Infermedica runs for symptom checking. The partner keeps the brand, the workflow, and the customer relationship. Digits Core Service provides the regulated hand-assessment capability they cannot economically build themselves.

Embedded inside

Telemedicine platform

What: Adds a hand ROM + dexterity step inside its existing video visit, returning structured Observations to the visit note.

Why: Hand triage is high-volume and hard to do over video; DCS turns the patient's phone into the goniometer.

Embedded inside

EHR vendor

What: Ships a hand-assessment order type; results flow back as DeviceMetric + Observation, attached to the encounter.

Why: The EHR keeps the chart and the buyer; DCS supplies the clinical capability without the EHR training models.

Embedded inside

Pharmacy / retail health

What: Embeds at-home arthritis tracking inside the existing pharmacy app; flags decline and offers a teleconsult.

Why: Pharmacies own daily patient touchpoints but not clinical measurement; DCS slots in as the measurement layer.

Embedded inside

OT / PT network

What: Pre-visit assessment, between-session homework, and outcome reports — all branded as the network's program.

Why: Therapists keep clinician trust and distribution; DCS handles the kinematics and the report.

Embedded inside

Trial sponsor / CRO

What: Calls the DCS API as an instrumented endpoint inside a hand-condition study; pulls FHIR Observations into the CTMS.

Why: Sponsors get a validated, decentralized endpoint without recruiting a research-imaging vendor.

Embedded inside

Hospital innovation team

What: Wraps DCS inside a department-branded follow-up app for post-op hand surgery patients.

Why: Hospitals optimize for workflow and procurement; DCS is the rentable intelligence the internal team cannot build.

What the partner gets when they integrate DCS

Each row is a surface that already exists in the Digits codebase today — DCS is the productized name for the bundle, not a rewrite.

DCS surfaceBacking codeWhat the partner can do
POST /api/assessments/romapps/api/src/routes/assessments.tsSubmit per-joint ROM frames, receive a validated assessment record + AI-ready summary.
POST /api/biomechanics/*apps/api/src/routes/biomechanics.tsDexterity (finger tapping, guided ROM), grip, tremor, swelling — symptom primitives.
WS /api/ai/wsapps/api/src/ai/orchestrator.tsVoice-guided assessment session; partner UI streams transcripts + receives state + TTS audio.
POST /api/reportsapps/api/src/routes/reports.tsGenerate a clinician-signable PDF with longitudinal trends, sharable by credential.
Capture SDK (web + Android)apps/web/src/lib/ hooks + apps/android/.../data/api/Drop-in MediaPipe-backed capture in the partner's UI; assessment state machine is exported.
Shared schemaspackages/sharedZod + TypeScript DTOs, Retrofit-mirrored on Android — one contract across every embed.
Outbox → substrateIntegration bridge (planned, see myLaminin post)Mirror every assessment as FHIR Observation + anchor signed reports to a regulated-research ledger.
Why now

The hyperscalers arrive — and the moat moves outward

Within a single quarter, OpenAI launched ChatGPT Health, Anthropic launched Claude for Healthcare, and AWS launched Amazon Connect Health. All three entered with compliant infrastructure, EHR connectivity, and enterprise customers attached to the launch. Generic compliant inference is becoming abundant fast. The floor of the market just rose.

That changes where defensibility lives. It is no longer the wrapper. It is the things that cannot be conjured by a frontier lab in ninety days:

1
Clinical validation depth
Years of video-grounded ROM data, joint-level error bounds, and condition-specific cohorts. The DRC pipeline is the engine that compounds this.
2
Workflow integration
Per-partner adapters into EHR encounters, telemedicine flows, pharmacy apps — each integration is its own moat.
3
Clinician trust
Reports clinicians actually sign. Outputs tiered as clinical-grade vs. patient-grade. No API shortcut around this.
4
Regulatory posture
Clearance scoped to the right output tier, consent + audit handled at the substrate layer, holder-controlled sharing.
What we are actually building

A bundle, not a model

The category still does not really have a clean name. Healthcare AI infrastructure, clinical AI APIs, AI as a service — none of those quite capture what makes this layer structurally different. The defining characteristic is the bundle: intelligence, validation, regulatory wrapper, and access surface, delivered together as something another company can directly embed. Sold once. Integrated once. Governed once.

That is what we mean when we say Digits Core Service is the Clinical API Layer for the hand. Not that we have an API — anyone can ship an API. That the hand is finally becoming programmable as a clinically trusted primitive, and Digits is positioned to be the substrate that the rest of healthcare rents when they need it.

The interface stays where it already is. The intelligence becomes modular. The defensibility moves into data, workflow, trust, and regulatory posture. Digits Core Service is the hand-rehab instance of that pattern.

Position paper. References the existing Digits codebase as it stands today and the DRC → DCS path memo for delivery sequencing.