Overview
Augmedix is the ambient AI documentation engine inside Commure. Founded in 2012 at Stanford on the premise of streaming a clinician’s first-person view to remote scribes in Dhaka, it evolved through Google Glass, hybrid AI-assisted scribing, and finally a fully automated multi-LLM stack on Google Cloud. Following the October 2024 take-private, Augmedix is rebranded under the “Commure Ambient” banner and serves as the documentation front end of Commure’s end-to-end clinical platform.
What it is, in one paragraph
An iOS app on the clinician’s phone (or a Bluetooth lab-coat mic) captures the encounter audio. Audio streams to Google Cloud, where Speech-to-Text produces a diarized transcript, then a multi-pass NLP/LLM pipeline on Vertex AI — using MedLM, MedGemma, Gemini, and Augmedix-fine-tuned models — drafts a clinical note across HPI, PE, Assessment & Plan. Depending on product tier, a Medical Documentation Specialist in Bangladesh or India polishes the draft inside NoteBuilder before it is written back to the EHR (Epic, Oracle Health, athenahealth, MEDITECH) as a FHIR DocumentReference. The clinician reviews and signs.
The trajectory in one paragraph
Augmedix’s architectural story is a gradient: from 100% human (Glass → scribe) in 2014, to AI-assisted scribes (NoteBuilder, 2021), to fully autonomous AI drafting (Go, September 2023), to ED-specialized AI (Go ED, April 2024), and finally to a deeply integrated layer in Commure’s revenue-cycle platform — where ambient notes feed autonomous coding, CDI, and claims submission. Across that arc the GCP substrate has remained constant: STT, Vertex AI, Pub/Sub, Spanner, BigQuery, FHIR API.
The Six Products
Augmedix’s portfolio is a gradient along a single axis: human-scribe-heavy on one end, fully autonomous AI on the other. All six products share the same NoteBuilder/Vertex AI substrate; what differs is how much human labor sits in the loop and where in the encounter the work happens. Post-acquisition all six are marketed under “Commure Ambient.”
graph LR
PREP["Augmedix Prep
(pre-visit chart
preparation, Sep 2022)"]
LIVE["Augmedix Live
(real-time human MDS,
2014 origin)"]
NOTES["Augmedix Notes /
NoteBuilder
(AI draft + MDS, 2021)"]
ASSIST["Augmedix Assist
(AI + 1-hr MDS polish)"]
GO["Augmedix Go
(fully automated AI,
Sep 2023)"]
GOED["Augmedix Go ED
(ED-specialized AI,
Apr 2024)"]
PREP --> LIVE
LIVE --> NOTES
NOTES --> ASSIST
ASSIST --> GO
GO --> GOED
style PREP fill:#c9b87a,stroke:#704214,color:#2a1f1a
style LIVE fill:#a8533a,stroke:#722f37,color:#f4e4c1
style NOTES fill:#8b6914,stroke:#4a3728,color:#f4e4c1
style ASSIST fill:#704214,stroke:#4a3728,color:#f4e4c1
style GO fill:#8b3a4a,stroke:#4a3728,color:#f4e4c1
style GOED fill:#722f37,stroke:#2a1f1a,color:#f4e4c1
Leftmost (Prep, Live) is the most human-labour-intensive: a Medical Documentation Specialist watches the encounter live or builds context ahead of it. Middle (Notes, Assist) is the hybrid tier: AI drafts, humans polish — the original NoteBuilder workflow. Rightmost (Go, Go ED) is fully autonomous: AI drafts the note within moments, the clinician reviews and signs, no human scribe in the loop.
| Product | Launched | Human in Loop? | Delivery Time | Claimed Savings |
|---|---|---|---|---|
| Augmedix Prep | Sep 2022 | MDS QA on pre-built note structure | Before visit | Chart-prep burden reduced |
| Augmedix Live | 2014 (Glass era) | MDS watches encounter in real time | Signature-ready immediately post-visit | 2–3 hrs/day |
| Augmedix Notes / NoteBuilder | Apr 2021 | MDS finalizes AI-assisted draft | Same day | Per-note labour cut |
| Augmedix Assist | ~2023 | MDS edits AI draft + coding | Within 1 hour | 1.5–2 hrs/day |
| Augmedix Go | Sep 27, 2023 | None — clinician reviews directly | Within moments | ~1 hr/day |
| Augmedix Go ED | Apr 24, 2024 | None — first fully automated ED ambient AI | Continuous across multi-touch ED visits | 99% patient consent in HCA pilot |
Go · The flagship
iOS app, iPhone 12+, iOS 15+. Multi-recording stitching used by 50% of clinicians. CareCues real-time care-gap alerts. ICD-10 suggestions. Problems Page. 32-language support. visionOS 1.0+ supported.
Go ED · The hardest case
Purpose-built for emergency departments: multi-hour, non-sequential interactions with one patient, high ambient noise, documenting reasoning across re-evaluations. Piloted at four HCA hospitals.
Live · The original
Highest-touch tier. MDS observes encounter, two-way chat with clinician, signature-ready note immediately. Includes prior-auth assistance, referral letters, AVS drafting, CPT/E&M/HCC coding.
NoteBuilder · The brain
The proprietary internal tool that all human-in-loop tiers run on. ASR + NLP surface dynamically filtered structured suggestions by section (HPI, ROS, PE, A&P) and specialty. Medication datasets integrated.
Prep · The pre-visit layer
Pre-populates patient demographics, med changes, history, labs/imaging, immunizations, family/social. Physician walks into the room with a pre-built note structure. McKesson Ventures highlighted as strategic.
Structured Data · The exhaust
Beyond notes, Augmedix exports three artefacts per encounter: full transcript, clinical note (pre and post physician edit), and structured clinical data — with optional PHI de-identification on the export path.
Capture Pipeline — End-to-End
A single clinical encounter traces a consistent path from the clinician’s pocket microphone to the signed EHR note. The specific hop count varies by product tier (Go skips the human-scribe queue), but the skeleton is the same: capture → transcribe → reason → (optionally polish) → write back → sign.
sequenceDiagram
participant C as Clinician
participant APP as iOS App
+ Bluetooth Mic
participant STT as Google STT
(diarization)
participant VAI as Vertex AI Pipeline
(MedLM + MedGemma + Gemini)
participant MDS as MDS Queue
(Live / Assist tiers)
participant NB as NoteBuilder QA
participant EHR as EHR Write-Back
(FHIR DocumentReference)
participant SIGN as Clinician Sign-Off
C->>APP: Speak with patient
APP->>STT: Stream audio
STT->>VAI: Diarized transcript
VAI->>VAI: Multi-pass section drafting
alt Live / Assist tier
VAI->>MDS: AI draft to scribe queue
MDS->>NB: Human polish in NoteBuilder
NB->>EHR: Upload pending note
else Go / Go ED tier
VAI->>EHR: AI draft direct
end
EHR->>SIGN: Note in pending status
SIGN->>EHR: Final signed note
graph LR
CLIN["Clinician"]
APP["iOS App
(Augmedix Go)"]
MIC["Bluetooth Mic
(lab coat pocket)"]
STT["Google Cloud
Speech-to-Text"]
NLP["NLP Extraction
(specialty + section)"]
LLM["Multi-LLM Pass
(MedLM / MedGemma /
Gemini + cross-check)"]
MDS["MDS Queue
(Bangladesh / India,
Live + Assist only)"]
NB["NoteBuilder QA
(structured suggestions,
med datasets)"]
EHR["EHR Write-Back
(FHIR DocumentReference
or section-level)"]
SIGN["Clinician
Reviews & Signs"]
CLIN --> APP
MIC --> APP
APP --> STT
STT --> NLP
NLP --> LLM
LLM --> MDS
LLM --> EHR
MDS --> NB
NB --> EHR
EHR --> SIGN
style CLIN fill:#5a4633,stroke:#2a1f1a,color:#f4e4c1
style APP fill:#6b8e6f,stroke:#2d4a3e,color:#2a1f1a
style MIC fill:#6b8e6f,stroke:#2d4a3e,color:#2a1f1a
style STT fill:#4a6b8a,stroke:#1e3a5f,color:#f4e4c1
style NLP fill:#8b3a4a,stroke:#722f37,color:#f4e4c1
style LLM fill:#8b3a4a,stroke:#722f37,color:#f4e4c1
style MDS fill:#a8533a,stroke:#722f37,color:#f4e4c1
style NB fill:#a8533a,stroke:#722f37,color:#f4e4c1
style EHR fill:#8b6914,stroke:#4a3728,color:#f4e4c1
style SIGN fill:#5a4633,stroke:#2a1f1a,color:#f4e4c1
Per the Google Cloud Developer Community article (June 2024), Augmedix retains encounter audio and note data for one week after EMR transmission before permanent deletion. This is shorter than the “up to six years after account closure” window stated in the general privacy policy — the one-week figure is specifically for the audio + raw transcript artefacts under the BAA. The shorter retention is part of the BAA boundary that lets identifiable PHI flow through Vertex AI inference.
The Go product schedule confirms that the app requires continuous internet connectivity during use; audio processing is server-side, not on-device. The Live schedule additionally requires WPA2 PSK (AES) or WPA2 PEAP wireless authentication on the clinical network with a minimum −65 dBm RSSI and QoS elevated to platinum/voice priority.
The Google Cloud Substrate
Augmedix is one of the deepest single-vendor Google Cloud deployments in healthcare AI. The richest public technical source — the Google Cloud Developer Community article from June 2024 — enumerates the services by layer: ASR/AI, messaging, storage, compute, analytics. SADA serves as the integration partner; Augmedix maintains a Google Cloud Marketplace listing so health systems can procure through GCP billing.
graph TD
subgraph CAPTURE["Capture Layer"]
IOS["iOS App
(Augmedix Go)"]
end
subgraph ASRAI["ASR & AI Layer"]
STT["Speech-to-Text
(diarization)"]
VAI["Vertex AI"]
MEDLM["MedLM
(Med-PaLM 2 lineage)"]
MEDG["MedGemma"]
GEM["Gemini
(multimodal)"]
end
subgraph MSG["Messaging Layer"]
PUBSUB["Cloud Pub/Sub
(encounter topics)"]
end
subgraph COMPUTE["Compute Layer"]
GKE["GKE
(encounter_service)"]
CRUN["Cloud Run"]
end
subgraph STORAGE["Storage Layer"]
SPAN["Cloud Spanner
(encounter tables)"]
BQ["BigQuery
(analytics warehouse)"]
HCFHIR["Healthcare FHIR API
(R4)"]
HDE["Healthcare Data
Engine (HDEv2)"]
end
subgraph ANALYTICS["Analytics Layer"]
LOOK["Looker
(dashboards)"]
COLAB["Vertex AI Workbench /
Colab Enterprise"]
end
IOS --> STT
STT --> VAI
VAI --> MEDLM
VAI --> MEDG
VAI --> GEM
VAI --> PUBSUB
PUBSUB --> GKE
GKE --> SPAN
SPAN --> BQ
HCFHIR --> HDE
HDE --> BQ
BQ --> LOOK
BQ --> COLAB
style IOS fill:#6b8e6f,stroke:#2d4a3e,color:#2a1f1a
style STT fill:#4a6b8a,stroke:#1e3a5f,color:#f4e4c1
style VAI fill:#8b3a4a,stroke:#722f37,color:#f4e4c1
style MEDLM fill:#8b3a4a,stroke:#722f37,color:#f4e4c1
style MEDG fill:#8b3a4a,stroke:#722f37,color:#f4e4c1
style GEM fill:#8b3a4a,stroke:#722f37,color:#f4e4c1
style PUBSUB fill:#c9b87a,stroke:#704214,color:#2a1f1a
style GKE fill:#704214,stroke:#4a3728,color:#f4e4c1
style CRUN fill:#704214,stroke:#4a3728,color:#f4e4c1
style SPAN fill:#4a6b8a,stroke:#1e3a5f,color:#f4e4c1
style BQ fill:#4a6b8a,stroke:#1e3a5f,color:#f4e4c1
style HCFHIR fill:#4a6b8a,stroke:#1e3a5f,color:#f4e4c1
style HDE fill:#4a6b8a,stroke:#1e3a5f,color:#f4e4c1
style LOOK fill:#8b6914,stroke:#4a3728,color:#f4e4c1
style COLAB fill:#8b6914,stroke:#4a3728,color:#f4e4c1
| Service | Layer | Role |
|---|---|---|
| Speech-to-Text | ASR | Diarized transcript from raw encounter audio |
| Vertex AI | AI platform | Hosts MedLM, MedGemma, Gemini; fine-tuning runs for Augmedix-proprietary models |
| MedLM on Vertex AI | Foundation models | Medical foundation models built on Med-PaLM 2 lineage; early-tester since Dec 2023 |
| MedGemma | Foundation models | Medical-specific Gemini-derivative; confirmed in Google Dev article as active in the Clinical Notes Engine |
| Gemini | Foundation models | Multimodal LLM supporting advanced note processing |
| Cloud Pub/Sub | Messaging | Asynchronous publish of structured encounter JSON to customer topics |
| GKE | Compute | Hosts encounter_service container that subscribes to Pub/Sub and writes to Spanner |
| Cloud Run | Compute | Alternative serverless deployment of encounter_service |
| Cloud Spanner | Storage | Relational encounter table storing structured clinical data |
| BigQuery | Storage / analytics | Combines Augmedix encounter data with longitudinal patient records for analytics |
| Healthcare FHIR API | Storage | FHIR R4 patient record store; RESTful patient data access |
| Healthcare Data Engine (HDEv2) | Storage / harmonization | Aggregates and standardizes multi-source data to FHIR R4 (private preview as of June 2024) |
| Looker | Analytics | Population dashboards, care-gap alerts, clinical recommendations |
| Vertex AI Workbench / Colab Enterprise | Analytics | Used by health-system AI researchers for custom model dev on the encounter corpus |
SADA · Integration Partner
SADA — a Google Cloud premier partner — was named (April 2024) as Augmedix’s scaling partner for ambient documentation and structured data products to health systems nationwide.
Google Cloud Marketplace
Augmedix maintains a Marketplace listing at console.cloud.google.com/marketplace/product/augmedix-public/augmedix, enabling health systems to procure Augmedix through GCP billing.
BAA Boundary
Google’s HIPAA BAA covers Vertex AI as a HIPAA-eligible service. Per the BAA, Vertex AI does not use customer data, prompts, or responses to improve or train the foundation models — the legal basis that lets identifiable PHI flow through inference.
Enterprise Data Lake Path
For enterprise customers, Augmedix publishes structured JSON to a customer-defined Pub/Sub topic; the customer’s encounter_service on GKE or Cloud Run subscribes and lands data in their own Spanner / BigQuery. Available within 30 minutes of EHR transmission. 99.0% uptime SLA.
EHR Integrations
Augmedix supports bidirectional integration with four named EHR platforms via two protocols: HL7 v2 (for schedule pulls) and FHIR R4 (for encounter context and DocumentReference write-back). Specifics vary by EHR. By the time of the Vizient contract (January 2025), Augmedix advertised “50+” EHR integrations; the named four below are the canonical relationships with public certification.
graph LR
AUGX["Augmedix
encounter_service"]
HL7["HL7 v2
(schedule pulls,
ADT/SIU feeds)"]
FHIR["FHIR R4
(encounter context,
DocumentReference)"]
EPIC["Epic
(Showroom Connection Hub,
listing 1103)"]
ORACLE["Oracle Health
(Validated Integration,
Dec 2024)"]
ATHENA["athenahealth
(athenaConnect,
section write-back)"]
MEDI["MEDITECH
(Expanse Now embedded,
Sept 2025)"]
AUGX --> HL7
AUGX --> FHIR
HL7 --> EPIC
FHIR --> EPIC
FHIR --> ORACLE
FHIR --> ATHENA
FHIR --> MEDI
style AUGX fill:#5a4633,stroke:#2a1f1a,color:#f4e4c1
style HL7 fill:#c9b87a,stroke:#704214,color:#2a1f1a
style FHIR fill:#c9b87a,stroke:#704214,color:#2a1f1a
style EPIC fill:#8b6914,stroke:#4a3728,color:#f4e4c1
style ORACLE fill:#8b6914,stroke:#4a3728,color:#f4e4c1
style ATHENA fill:#8b6914,stroke:#4a3728,color:#f4e4c1
style MEDI fill:#8b6914,stroke:#4a3728,color:#f4e4c1
| EHR | Pull Protocol | Write-Back Mechanism | Certification |
|---|---|---|---|
| Epic | HL7 v2 schedules + FHIR R4 encounter context | FHIR DocumentReference / SmartText / Smart Data Elements inside Hyperspace/Hyperdrive | Epic Showroom Connection Hub tier — listing id 1103 ($500/yr) |
| Oracle Health (Cerner) | HL7 + FHIR | Validated cross all three Augmedix product tiers (Go, Assist, Live) | Oracle Validated Integration — December 2024; 10+ year relationship |
| athenahealth | FHIR APIs (schedule + encounter) | Section-level write-back: HPI / ROS / PE / A&P into encounter template fields | Listed in athenaConnect marketplace |
| MEDITECH | Bidirectional; specific protocol not publicly documented (FHIR R4 inferred) | Commure Ambient embedded natively in MEDITECH Expanse Now (mobile) | Commure joined MEDITECH Alliance Mar 2025; Expanse Now embed live Sept 2025 |
When the Augmedix app launches inside Epic Hyperspace/Hyperdrive, it receives the encounter ID, patient ID, and clinician ID via the Epic-issued SMART launch token, along with OAuth 2.0 authorization scopes. The clinician authenticates once to Epic (commonly through Imprivata-backed IdP); Augmedix inherits the authenticated context. No separate Augmedix login required in this flow.
When EHR integration cannot be configured (for organizational or technical reasons), clinicians can manually copy-paste notes from the Augmedix Customer Portal into their EHR. This degrades the feature set: no automatic structured data upload, no pre-population of pre-visit context. The Go product schedule names this explicitly as a supported degraded mode.
The Multi-Pass, Multi-LLM Pipeline
Augmedix’s key architectural differentiator from single-LLM ambient-scribe competitors is its multi-model, multi-pass design: section-specific specialized models drafting different parts of the note (HPI, PE, A&P), with cross-checking between models to instill clinical accuracy and physician trust. The system was an early-tester of MedLM at its December 2023 launch and integrates Augmedix-fine-tuned models on the same Vertex AI substrate.
graph TD
AUDIO["Encounter Audio"]
STT["Google STT
(diarized transcript)"]
CTX["Context Retrieval
(prior notes, problem list,
meds via FHIR)"]
subgraph PASS["Section-Specific LLM Passes"]
HPI["HPI Pass
(history of
present illness)"]
PE["PE Pass
(physical exam)"]
AP["A&P Pass
(assessment + plan)"]
ROS["ROS Pass
(review of systems)"]
end
XCHECK["Cross-Check Pass
(model-vs-model
quality assurance)"]
STRUCT["Structured Data
Extraction
(ICD-10, meds, labs)"]
ASSEMBLY["Note Assembly
(section stitching)"]
QA["NoteBuilder QA
(human or AI tier)"]
AUDIO --> STT
STT --> CTX
CTX --> HPI
CTX --> PE
CTX --> AP
CTX --> ROS
HPI --> XCHECK
PE --> XCHECK
AP --> XCHECK
ROS --> XCHECK
XCHECK --> ASSEMBLY
XCHECK --> STRUCT
ASSEMBLY --> QA
style AUDIO fill:#6b8e6f,stroke:#2d4a3e,color:#2a1f1a
style STT fill:#4a6b8a,stroke:#1e3a5f,color:#f4e4c1
style CTX fill:#4a6b8a,stroke:#1e3a5f,color:#f4e4c1
style HPI fill:#8b3a4a,stroke:#722f37,color:#f4e4c1
style PE fill:#8b3a4a,stroke:#722f37,color:#f4e4c1
style AP fill:#8b3a4a,stroke:#722f37,color:#f4e4c1
style ROS fill:#8b3a4a,stroke:#722f37,color:#f4e4c1
style XCHECK fill:#722f37,stroke:#2a1f1a,color:#f4e4c1
style STRUCT fill:#c9b87a,stroke:#704214,color:#2a1f1a
style ASSEMBLY fill:#704214,stroke:#4a3728,color:#f4e4c1
style QA fill:#a8533a,stroke:#722f37,color:#f4e4c1
graph LR
NOTES["70K Notes/Week
Production Corpus"]
SME["Specialty SMEs
(rubric scoring,
50+ specialties)"]
QUAL["Quality Scoring
(internal rubrics)"]
TUNE["Vertex AI
Fine-Tuning Job"]
BASE["MedLM Base /
MedGemma / Gemini"]
SPEC["Specialty-Specific
Fine-Tuned Models"]
DEPLOY["Vertex AI
Endpoint Deploy"]
PROD["Production Inference
(Go / Assist / Live)"]
NOTES --> SME
SME --> QUAL
QUAL --> TUNE
BASE --> TUNE
TUNE --> SPEC
SPEC --> DEPLOY
DEPLOY --> PROD
PROD --> NOTES
style NOTES fill:#c9b87a,stroke:#704214,color:#2a1f1a
style SME fill:#a8533a,stroke:#722f37,color:#f4e4c1
style QUAL fill:#a8533a,stroke:#722f37,color:#f4e4c1
style TUNE fill:#8b3a4a,stroke:#722f37,color:#f4e4c1
style BASE fill:#8b3a4a,stroke:#722f37,color:#f4e4c1
style SPEC fill:#722f37,stroke:#2a1f1a,color:#f4e4c1
style DEPLOY fill:#4a6b8a,stroke:#1e3a5f,color:#f4e4c1
style PROD fill:#5a4633,stroke:#2a1f1a,color:#f4e4c1
MedLM · The early bet
Announced December 11, 2023 by Google. Built on Med-PaLM 2 (which scored 86.5% on US medical licensing exam questions). Augmedix was an early tester alongside HCA Healthcare and BenchSci. Now integrated across all Augmedix products via Vertex AI inference.
MedGemma · The next generation
Google’s subsequent medical foundation model, based on Gemini rather than PaLM. Confirmed in the June 2024 Google Dev article as actively handling healthcare-specific language understanding in the Augmedix pipeline.
Gemini · The multimodal layer
Underpins multimodal processing in the stack. Where audio characteristics, room context, or other non-text signals become relevant, Gemini provides the multimodal LLM backbone.
Augmedix proprietary models
Specialty-specific fine-tuned models trained on Augmedix’s 70K-notes/week corpus across 50+ specialties. Quality is scored against internal rubrics with feedback from human SMEs. This is the loop that turns operational scale into model quality.
Cross-check pass
Augmedix’s own framing: “By utilizing multiple models that cross-check each other, Augmedix is delivering high quality notes that instill trust among clinicians.” A second model reviews a first model’s output before assembly.
Structured-data extraction
A separate NLP path generates structured outputs — ICD-10 codes, medications with dosage/frequency, labs, A&P entries — in parallel with the narrative note. These feed both EHR discrete fields and the BigQuery analytics warehouse.
The HCA Healthcare pilot reported a clinical documentation F1 score > 0.8 with 1,400+ physicians piloting across 50+ hospitals. This indicates clinical note accuracy is competitive but not yet signature-ready for every specialty — which is why the Assist and Live human-review tiers continue to exist alongside fully autonomous Go.
Human in the Loop
The Medical Documentation Specialist (MDS) workforce — centred in Dhaka, with additional vendor capacity in India — is the operational story behind Augmedix’s reputation for quality. At peak (pre-acquisition) the Bangladesh subsidiary employed 800+ staff including 500+ scribes, operating from HIPAA-secure rooms with three-month paid training programs. The architectural question is not “is there a human?” but “where in the workflow does the human sit?” — and that answer differs by product tier.
flowchart LR
ENC["Encounter Audio
+ AI Draft"]
ROUTER{"Workflow Router
(by product tier)"}
LIVEPATH["Live Tier
100% Human Review
(real-time synchronous)"]
NOTESPATH["Notes / Assist Tier
AI Drafts +
MDS Polish"]
GOPATH["Go / Go ED Tier
AI Only +
Physician Sign-Off"]
MDSLIVE["MDS in HIPAA Room
(watches encounter live,
NoteBuilder)"]
MDSNOTES["MDS Reviews Draft
(NoteBuilder + coding,
within 1 hour)"]
PHYS["Physician
Review + Sign"]
EHRPEND["EHR
Pending Status"]
EHRSIGN["EHR
Final Signed"]
ENC --> ROUTER
ROUTER --> LIVEPATH
ROUTER --> NOTESPATH
ROUTER --> GOPATH
LIVEPATH --> MDSLIVE
NOTESPATH --> MDSNOTES
GOPATH --> PHYS
MDSLIVE --> EHRPEND
MDSNOTES --> EHRPEND
PHYS --> EHRSIGN
EHRPEND --> EHRSIGN
style ENC fill:#6b8e6f,stroke:#2d4a3e,color:#2a1f1a
style ROUTER fill:#704214,stroke:#4a3728,color:#f4e4c1
style LIVEPATH fill:#a8533a,stroke:#722f37,color:#f4e4c1
style NOTESPATH fill:#a8533a,stroke:#722f37,color:#f4e4c1
style GOPATH fill:#8b3a4a,stroke:#722f37,color:#f4e4c1
style MDSLIVE fill:#a8533a,stroke:#722f37,color:#f4e4c1
style MDSNOTES fill:#a8533a,stroke:#722f37,color:#f4e4c1
style PHYS fill:#5a4633,stroke:#2a1f1a,color:#f4e4c1
style EHRPEND fill:#8b6914,stroke:#4a3728,color:#f4e4c1
style EHRSIGN fill:#8b6914,stroke:#4a3728,color:#f4e4c1
HIPAA does not prohibit offshore PHI access, provided the Business Associate Agreement covers the subcontractor and appropriate technical and physical safeguards are in place. Augmedix’s BAA with each hospital covers its scribe operations as a downstream subcontractor relationship; MDSs operate from HIPAA-secure rooms with controlled access, no personal devices, and clean-desk policies. MDSs receive credentials directly into the customer’s EHR — meaning they upload pending notes under the same credential governance as the host health system’s own staff.
NoteBuilder · The MDS workstation
The internal tool every human-tier scribe uses. Dynamic templates filtered by specialty, visit type, and complaint. Intelligent auto-completion narrows options as the scribe types. Integrated medication datasets with dosages and frequencies. Real-time AI assistance generating HPI, PE, A&P sections.
Bangladesh primary site
Augmedix Bangladesh: 800+ total staff at peak, 500+ scribes, three-month paid MDS training program. Company leadership signaled plans for a 7,000-person Bangladesh workforce over 5 years (as of ~2022).
India vendor capacity
India scribes operate through vendor relationships rather than a direct subsidiary. At one point approximately 70% of Remote Documentation Specialists resided in India alongside the Bangladesh subsidiary’s direct employees.
Five-country footprint
US (San Francisco HQ), Bangladesh (Dhaka, primary), India, Sri Lanka, Dominican Republic. The geographic spread provides 24-hour follow-the-sun coverage for Live encounters across US time zones.
Downstream BAA chain
The legal structure: Hospital signs BAA with Augmedix → Augmedix flows the BAA terms down to its offshore subsidiary and vendor scribes. MDSs are bound by the same PHI handling rules as on-shore Business Associate staff. Augmedix’s HITRUST r2 covers this chain.
Two-way clinician chat
In Live, the MDS and clinician communicate via text during the encounter — the MDS can ask for a clarification, the clinician can flag an order to add. This is the operational mechanism that lets a remote scribe stay in sync with a fast-moving in-room visit.
The transition story
The deliberate reduction in per-note human labour is the central architectural story of Augmedix’s ten years:
- 2013–2021 · Google Glass + full human scribe. Scribes watched the entire encounter through the clinician’s point-of-view stream and wrote the note from scratch. AI minimal. Value was human labour arbitrage.
- 2021–2023 · NoteBuilder AI assistance + human scribe. AI pre-drafts sections; scribes correct and finalize. The S-1 explicitly called out NoteBuilder as central to the plan to “reduce the level of effort required by remote documentation specialists.”
- 2023–present · Go (pure AI) + Assist (human polish on AI draft) + Live (real-time human). The Go tier represents the shift from human-first to AI-first documentation. Partly accelerated by COVID-era constraints that made in-person scribing infeasible.
Per Commure’s acquisition press release, all three tiers continue to receive investment post-acquisition — the architecture isn’t “replace humans”, it’s “tier humans by where they add the most value relative to the AI baseline.”
Scribe Operations — A Closer Look
The scribe operation is the human substrate that has carried Augmedix for a decade. Encounter audio leaves the clinician’s phone, gets routed by tier, and lands in a queue watched by a Medical Documentation Specialist (MDS) sitting in a HIPAA-secure room — mostly in Dhaka, with vendor capacity in India and follow-the-sun coverage from Sri Lanka and the Dominican Republic. NoteBuilder is the workstation. The strategic arc has been steady: more AI in the loop, less labour per note — without dismantling the human review tier, because Live + Assist are precisely what differentiate Augmedix from pure-AI competitors.
graph TD
AUDIO["Encounter Audio
(iOS app / Bluetooth mic)"]
ROUTE{"Tier Router"}
LIVEQ["Live Queue
(synchronous,
follow-the-sun)"]
ASSISTQ["Assist Queue
(post-encounter,
1-hour SLA)"]
GOSKIP["Go · No Human"]
subgraph SCRIBE_GEO["Scribe Geographies (HIPAA-secure rooms)"]
BD["Bangladesh
(Dhaka subsidiary,
500+ MDSs)"]
IN["India
(vendor partners,
~70% of RDSs)"]
LK["Sri Lanka
(coverage)"]
DR["Dominican Republic
(coverage)"]
end
NB["NoteBuilder
(MDS workstation)"]
QA["QA · Quality Review
(internal rubric)"]
EHR_WRITE["EHR Write-Back
(MDS-credentialed,
pending status)"]
PHYS_SIGN["Physician Review & Sign"]
AUDIO --> ROUTE
ROUTE --> LIVEQ
ROUTE --> ASSISTQ
ROUTE --> GOSKIP
LIVEQ --> BD
LIVEQ --> IN
LIVEQ --> LK
LIVEQ --> DR
ASSISTQ --> BD
ASSISTQ --> IN
BD --> NB
IN --> NB
LK --> NB
DR --> NB
NB --> QA
QA --> EHR_WRITE
GOSKIP --> EHR_WRITE
EHR_WRITE --> PHYS_SIGN
style AUDIO fill:#6b8e6f,stroke:#2d4a3e,color:#2a1f1a
style ROUTE fill:#704214,stroke:#4a3728,color:#f4e4c1
style LIVEQ fill:#a8533a,stroke:#722f37,color:#f4e4c1
style ASSISTQ fill:#a8533a,stroke:#722f37,color:#f4e4c1
style GOSKIP fill:#8b3a4a,stroke:#722f37,color:#f4e4c1
style BD fill:#c9b87a,stroke:#704214,color:#2a1f1a
style IN fill:#c9b87a,stroke:#704214,color:#2a1f1a
style LK fill:#c9b87a,stroke:#704214,color:#2a1f1a
style DR fill:#c9b87a,stroke:#704214,color:#2a1f1a
style NB fill:#6b4423,stroke:#4a3728,color:#f4e4c1
style QA fill:#6b4423,stroke:#4a3728,color:#f4e4c1
style EHR_WRITE fill:#8b6914,stroke:#4a3728,color:#f4e4c1
style PHYS_SIGN fill:#5a4633,stroke:#2a1f1a,color:#f4e4c1
HIPAA-secure rooms
MDSs operate from dedicated facilities with controlled access, no personal devices, and clean-desk policies. The room is the physical-safeguard layer that lets PHI flow across borders under the BAA chain.
NoteBuilder · The MDS IDE
Dynamic templates filtered by specialty, visit type, and complaint. Intelligent auto-completion narrows options as the scribe types. Integrated medication datasets (dosage, frequency, side effects). Real-time AI assistance drafting HPI, ROS, PE, A&P.
Downstream BAA chain
Hospital signs BAA with Augmedix → Augmedix flows BAA terms down to its Dhaka subsidiary and India vendor MDSs. Direct-EHR-access scribes are bound by the same PHI-handling rules as on-shore Business Associates.
COVID pivot to remote
In-person scribing became infeasible in 2020. The Bangladesh secure-room model was already remote by design; the constraint accelerated the smartphone capture push that ultimately replaced Google Glass and laid the groundwork for Go (pure-AI) two years later.
QA & calibration loop
MDS output feeds a continuous quality rubric: discrepancies between AI drafts and human polish become training signal for the next fine-tune. The S-1 explicitly framed NoteBuilder as “reducing the level of effort required by remote documentation specialists.”
From labour arbitrage to AI augmentation
2013–2021 the value was wage arbitrage (US physicians, Bangladesh scribes). 2021–2023 the value was hybrid AI+human. 2023+ the value is tiered: pure AI (Go), AI + human polish (Assist), AI + real-time human (Live). Same workforce, different leverage.
Compliance & PHI Handling
Augmedix’s compliance posture rests on three layered BAAs: Google Cloud’s BAA covering Vertex AI, Speech-to-Text, GKE, Spanner, BigQuery, Pub/Sub, and the Healthcare API; Augmedix’s BAA with each covered-entity customer; and downstream BAAs covering the offshore MDS subcontractors. HITRUST r2 (the most rigorous HITRUST tier) was earned August 2023 and explicitly scopes Augmedix Live, Augmedix Notes, and the supporting cloud services. There are no documented data breaches, no OCR enforcement actions, and no FTC matters. SOC 2 is not publicly confirmed.
| Certification / Posture | Date / Scope | Notes |
|---|---|---|
| HITRUST r2 Certified | August 2023 · 2-year | Scope: Augmedix Live, Augmedix Notes, supporting cloud services. The risk-based, two-year tier — HITRUST’s most rigorous, with independent assessor verification. |
| HIPAA BAA · Google Cloud | Ongoing | Covers Vertex AI, Speech-to-Text, GKE, Cloud Spanner, BigQuery, Pub/Sub, Cloud Storage, Healthcare API. The legal basis for identifiable PHI flowing through MedLM/MedGemma inference. |
| HIPAA BAA · Customer | Per-customer | Each hospital / health-system customer signs a BAA with Augmedix. Standard covered-entity ↔ business-associate structure. |
| Downstream BAAs · MDSs | Per-relationship | Augmedix Bangladesh subsidiary and India vendors flow down the BAA terms. Direct-EHR-access scribes are bound under the same credential governance. |
| ISO 27001 / 27017 / 27018 | GCP infra layer | Applies to Google Cloud substrate. Augmedix application-layer ISO posture not publicly documented separately from HITRUST. |
| SOC 2 | [unconfirmed] | Not mentioned in any public Augmedix source. HITRUST r2 is positioned as the company’s primary third-party assurance. |
| GDPR | Not applicable (US-only) | No evidence of EU deployments. Customer base is US health systems. |
| Documented breaches | None found | No reported Augmedix breach in HIPAA Journal databases or public sources as of May 2026. |
| FTC actions | None found | No FTC enforcement matters or consent orders documented. |
flowchart LR
CLIN["Clinician
(Covered Entity staff)"]
APP["Augmedix App
(iOS / Bluetooth mic)"]
subgraph BAA_GCP["GCP HIPAA BAA"]
STT2["Speech-to-Text"]
VAI2["Vertex AI
(MedLM, MedGemma,
Gemini)"]
STORE["Spanner / BigQuery /
Healthcare API"]
end
subgraph BAA_AUG["Augmedix-MDS BAA (downstream)"]
MDS2["MDS
(NoteBuilder,
HIPAA-secure room)"]
end
subgraph BAA_HOSP["Hospital BAA · EHR Tenant"]
EHR2["Epic / Oracle Health /
athenahealth / MEDITECH"]
end
CLIN --> APP
APP --> STT2
STT2 --> VAI2
VAI2 --> STORE
STORE --> MDS2
MDS2 --> EHR2
VAI2 --> EHR2
style CLIN fill:#5a4633,stroke:#2a1f1a,color:#f4e4c1
style APP fill:#6b8e6f,stroke:#2d4a3e,color:#2a1f1a
style STT2 fill:#4a6b8a,stroke:#1e3a5f,color:#f4e4c1
style VAI2 fill:#8b3a4a,stroke:#722f37,color:#f4e4c1
style STORE fill:#4a6b8a,stroke:#1e3a5f,color:#f4e4c1
style MDS2 fill:#a8533a,stroke:#722f37,color:#f4e4c1
style EHR2 fill:#8b6914,stroke:#4a3728,color:#f4e4c1
Google’s HIPAA BAA covers Vertex AI as a HIPAA-eligible service — the contract commits that Vertex AI does not use customer data, prompts, or responses to improve or train Google’s foundation models. This is what lets identifiable patient audio, transcripts, and notes flow through MedLM and MedGemma inference inside Augmedix’s pipeline without violating the minimum-necessary rule. Without that BAA scope, the same architecture would require de-identification before inference.
Patient consent
Customers are contractually required to obtain patient consent for audio recording “to the extent required under applicable law” — state wiretapping laws vary (one-party vs. two-party consent). HCA ED pilot reported 99% patient consent acceptance.
Audio retention
Per the Google Developer article: encounter audio and note data retained one week after EMR transmission, then permanently deleted. The general privacy policy speaks to “up to six years after account closure” for user account data — the audio window is narrower. [the one-week figure is from a blog, not a formal data-retention policy; treat as unconfirmed for audio specifically]
PHI de-identification path
Optional de-identified structured-data export for analytics customers whose data-lake use cases don’t require PHI. Mechanism (Cloud DLP, custom NER, or manual review) is not specified [unconfirmed].
Encryption
HITRUST r2 baseline implies AES-256 at rest, TLS 1.2+ in transit. The privacy policy speaks to “authentication, encryption, backups, and access controls” without naming protocols. CMEK / Confidential VMs / VPC Service Controls usage not publicly documented [unconfirmed].
Multi-Tenant Isolation on GCP
Augmedix’s multi-tenancy story is partly documented and partly inferred. The enterprise data-integration tier is unambiguous: Augmedix publishes encounter JSON into a customer-owned Pub/Sub topic that lands in a customer-owned encounter_service on the customer’s GCP project — a bring-your-own-project pattern. The inference pipeline (STT, Vertex AI, audio buffers) runs in Augmedix’s own GCP estate, where logical isolation is via IAM, encounter identifiers, and dataset-level permissions in BigQuery. The HITRUST r2 assessor reviewed the access-control domain, which means tenant separation in shared infrastructure was independently validated — but the specific control surface (per-tenant CMEK, VPC SC, project structure) is not published.
graph TB
subgraph AUG_PROJ["Augmedix-owned GCP Project (shared inference plane)"]
STT3["Speech-to-Text"]
VAI3["Vertex AI
(MedLM / MedGemma)"]
ENC_SVC["encounter_service
(per-tenant IAM)"]
SPAN3["Spanner
(encounter table,
tenant key in row)"]
end
subgraph TEN_A["Customer A · Their GCP Project"]
PSA["Pub/Sub topic A"]
SVCA["encounter_service A
(GKE / Cloud Run)"]
BQA["BigQuery A
(customer-owned)"]
end
subgraph TEN_B["Customer B · Their GCP Project"]
PSB["Pub/Sub topic B"]
SVCB["encounter_service B
(GKE / Cloud Run)"]
BQB["BigQuery B
(customer-owned)"]
end
IAM["Cloud IAM · per-tenant
service accounts &
dataset permissions"]
AUDIT["Cloud Audit Logs
[unconfirmed scope]"]
STT3 --> VAI3
VAI3 --> ENC_SVC
ENC_SVC --> SPAN3
ENC_SVC --> PSA
ENC_SVC --> PSB
PSA --> SVCA
PSB --> SVCB
SVCA --> BQA
SVCB --> BQB
IAM -.-> ENC_SVC
IAM -.-> PSA
IAM -.-> PSB
AUDIT -.-> ENC_SVC
style STT3 fill:#4a6b8a,stroke:#1e3a5f,color:#f4e4c1
style VAI3 fill:#8b3a4a,stroke:#722f37,color:#f4e4c1
style ENC_SVC fill:#704214,stroke:#4a3728,color:#f4e4c1
style SPAN3 fill:#4a6b8a,stroke:#1e3a5f,color:#f4e4c1
style PSA fill:#c9b87a,stroke:#704214,color:#2a1f1a
style PSB fill:#c9b87a,stroke:#704214,color:#2a1f1a
style SVCA fill:#704214,stroke:#4a3728,color:#f4e4c1
style SVCB fill:#704214,stroke:#4a3728,color:#f4e4c1
style BQA fill:#4a6b8a,stroke:#1e3a5f,color:#f4e4c1
style BQB fill:#4a6b8a,stroke:#1e3a5f,color:#f4e4c1
style IAM fill:#8b6914,stroke:#4a3728,color:#f4e4c1
style AUDIT fill:#6b4423,stroke:#4a3728,color:#f4e4c1
| Aspect | Confirmed / Inferred | Detail |
|---|---|---|
| Tenant unit | Confirmed | Health system / hospital. Each tenant is one BAA-signed customer; tenants are not per-clinician or per-encounter. |
| Enterprise data path | Confirmed (Google Dev blog) | Bring-your-own-project: customer-owned Pub/Sub topic, encounter_service, Spanner/BigQuery in the customer’s GCP estate. Augmedix publishes; customer owns the lake. |
| Inference pipeline isolation | Inferred | Shared Augmedix-owned GCP infrastructure (STT, Vertex AI, audio buffers) with logical isolation via IAM service accounts, encounter IDs (_encounterID, _noteID), and dataset-level BigQuery permissions. [unconfirmed specifics] |
| Spanner partitioning | Inferred | Encounter rows carry tenant identifiers (encounter ID, note ID, timestamps); per-tenant IAM enforces read scope. [unconfirmed schema] |
| CMEK (per-tenant keys) | Unconfirmed | Not publicly documented. Standard GCP-healthcare best practice supports CMEK via Cloud KMS; whether Augmedix offers per-tenant CMEK is not disclosed. |
| VPC Service Controls | Unconfirmed | GCP Healthcare best-practice pattern: wrap Vertex AI, Healthcare API, BigQuery in a service perimeter. Augmedix has not published its perimeter posture. |
| Audit logging | Implied | HIPAA requires; HITRUST r2 covers. Specific architecture (Cloud Logging exports, SIEM integration) not disclosed [unconfirmed]. |
| Tenant credential model | Confirmed | Each customer onboarded with separate EHR credentials and Augmedix account credentials. MDSs with EHR write access operate under the host health system’s own credential governance. |
The most architecturally interesting feature is the hybrid tenancy at the data-lake boundary: the inference plane is multi-tenant under Augmedix’s IAM, but the analytics plane is single-tenant inside each customer’s GCP project. This pushes the “data exfiltration boundary” out of Augmedix’s estate and into the customer’s — reducing Augmedix’s blast radius and giving each customer full control over their longitudinal patient record store. It also means BigQuery scale, costs, and governance live with the customer, not Augmedix.
The Commure Era
In July 2024 Commure — the General-Catalyst-backed healthcare-AI roll-up — signed a definitive agreement to take Augmedix private at $2.35/share, a $139M all-cash deal at a 169% premium to the 30-day VWAP. Stockholders approved September 27; closing was October 2, 2024. Augmedix delisted from Nasdaq (ticker AUGX) and became a wholly-owned Commure subsidiary. By mid-2025 Commure had raised $200M from General Catalyst’s Customer Value Fund; in May 2026 a $70M round closed at a $7B post-money valuation. The Augmedix ambient AI is now the front end of Commure’s end-to-end clinical platform — rebranded “Commure Ambient.”
timeline
title Augmedix → Commure · The Acquisition Arc
2012 : Founded at Stanford
: Glass + remote scribes thesis
2014 : Commercial launch
: Dignity / Sutter / CHI customers
2020 : Reverse-merger IPO
: Nasdaq AUGX · $40M raise
2023 : Glass deprecation
: MedLM early-tester (Dec)
: Augmedix Go launches (Sept)
2024 : Commure agreement (Jul)
: Take-private closes (Oct 2)
: $139M · 169% premium
2025 : Vizient contract (Jan)
: MEDITECH Alliance (Mar)
: Expanse Now embed (Sept)
2026 : Commure $7B valuation (May)
: Brand — Commure Ambient
What Commure brings to the architecture
Commure’s strategic thesis is vertical integration of the clinical revenue cycle: ambient documentation generates the note, autonomous coding codes it, CDI improves it, RCM bills it, and an agentic layer automates referrals, scheduling, and prior auth from the same clinical context. Augmedix supplies the documentation layer at the front of that stack. The post-acquisition footprint is materially larger than Augmedix achieved as a standalone public company.
HCA Healthcare · Exclusive partner
October 2024 press release named HCA as exclusive development partner for the multi-specialty ambient platform. 1,400+ physicians piloting across 50+ hospitals, clinical documentation F1 score > 0.8. Ian Shakil leads the HCA integration as Chief Strategy Officer at Commure.
Vizient contract
Awarded January 2025. Vizient is the healthcare GPO serving 65%+ of US acute-care providers, 97% of academic medical centers, 35% of non-acute, and $140B+ annual purchasing. Covers Go, Assist, and Live tiers — a distribution lever Augmedix never had alone.
MEDITECH Expanse Now embed
September 2025: Commure Ambient AI embedded directly in MEDITECH’s Expanse Now mobile EHR. Notes flow into discrete EHR sections across ED, inpatient, and ambulatory contexts — no copy-paste. Commure joined the MEDITECH Alliance in March 2025.
The integration thesis
Commure CEO Tanay Tandon framed it as “language models that transcribe appointments, autonomously code them, and supercharge back-office operations for billing teams.” Augmedix is the transcribe step; Athelas (acquired Oct 2023) is the code step; Commure Agents (launched June 2025) is the automation layer.
Commure platform scale
By 2025: 130+ health systems, 350,000+ clinicians, 60+ EHR integrations, $25B+ annual claims processed, 3,000+ sites of care, 40M+ ambient appointments / year. KLAS Research named Commure 2025 Market Leader in Ambient AI.
Capital trajectory
Oct 2023: Commure+Athelas merger, $70M from General Catalyst at ~$6B combined valuation. Oct 2024: Augmedix acquisition $139M. Dec 2024: Memora Health acquisition. Mid-2025: $200M from GC’s Customer Value Fund. May 2026: $70M round — $7B post-money with Sequoia, Morgan Stanley, Kirkland & Ellis participating.
Augmedix was growing (~40% YoY revenue in Q1 2024) but had never reached profitability and had cut its full-year 2024 revenue guidance to $52–55M, sending shares down 40%. The 169% premium over the 30-day VWAP indicates Commure valued the technology, the GCP/MedLM partnership, the offshore scribe workforce, and the HCA + customer relationships well above the depressed public-market price. From Commure’s side, owning the ambient documentation layer was the precondition for the vertically-integrated thesis — you cannot autonomously code a note you don’t own.
AI Model Lineage · The Scribe-to-LLM Arc
Chapter XI traced the corporate timeline. This one cuts the same period along a different axis: how the AI substrate evolved from minimal NLP layered onto Google Glass + human scribes, through proprietary specialty fine-tunes inside NoteBuilder, to a multi-LLM Vertex AI pipeline running MedLM, MedGemma, and Gemini in parallel passes. The architectural inflection is 2022–2023: capable LLMs make Go (pure-AI) viable, and the scribe role shifts from primary author to QA reviewer.
timeline
title Capture Modality + Model Lineage
2012 : Founding · Google Glass + remote scribes
: Minimal NLP, human-authored notes
2018 : Smartphone capture added
: Bluetooth lab-coat mic option
2020 : IPO (Nasdaq AUGX)
: Audio + ASR mature
2021 : NoteBuilder launches
: ASR + NLP + suggestion templates
2022 : Audio AI work scaled
: Specialty fine-tunes
: Augmedix Prep launches (Sept)
2023 : Glass sunset (Mar)
: MedLM early-tester (Dec)
: Augmedix Go launches (Sept)
2024 : MedGemma + Gemini in pipeline
: Go ED launches (Apr)
: Multi-LLM cross-check live
2025 : MEDITECH Expanse Now embed (Sept)
: Commure Agents adjacent
2026 : Commure Ambient brand
: Vertex AI / Healthcare API substrate
Three architectural inflections
- 2021 · NoteBuilder. First “AI in the loop” stage. ASR + NLP surfaced suggestions; humans still authored. The S-1 framed it as a labour-reduction tool, not a replacement.
- 2023 · The LLM inflection. Capable LLMs (Med-PaLM 2 lineage → MedLM) made Go viable. The role of the human shifted from primary author to QA reviewer (Assist) or was removed entirely (Go). HCA’s $12M equity investment in April 2023 funded this pipeline.
- 2024 · Multi-LLM cross-checking. The Google Developer article confirms MedLM + MedGemma + Gemini running together, with specialty fine-tunes and an explicit cross-check pass. This is Augmedix’s differentiator versus single-LLM ambient scribe competitors (Nuance DAX, Suki, Nabla, Abridge, Freed).
Acronyms & Sources
A glossary of every acronym used in this map — from the clinical (HPI, ROS, A&P) through the regulatory (HIPAA, HITRUST, BAA, PHI) to the GCP-architectural (GKE, FHIR, HL7). Followed by a consolidated source list and a brief provenance note.
AI
Artificial Intelligence — here, the umbrella for ASR, NLP, and LLM components in Augmedix’s pipeline.
A&P
Assessment & Plan — the section of a clinical note that records the clinician’s diagnostic reasoning and next steps.
ASR
Automatic Speech Recognition — converts spoken audio into text. Augmedix uses Google Cloud Speech-to-Text.
AVS
After-Visit Summary — the patient-facing document summarizing the encounter, instructions, and follow-ups.
BAA
Business Associate Agreement — the HIPAA contract between a Covered Entity (hospital) and a Business Associate (Augmedix, Google Cloud, MDSs) covering PHI handling.
BQ · BigQuery
Google Cloud’s serverless analytics data warehouse. Used here for longitudinal patient analytics combining Augmedix encounter data with EHR data.
CDI
Clinical Documentation Improvement — the process of refining clinical notes to support accurate coding and reimbursement.
CMEK
Customer-Managed Encryption Keys — GCP feature allowing customers to control encryption keys via Cloud KMS. Usage at Augmedix is unconfirmed.
CPT
Current Procedural Terminology — the AMA-maintained code set for medical procedures used in billing.
DLP
Data Loss Prevention — Google Cloud DLP is the de-identification / sensitive-data-detection service referenced as a complementary PHI scrubbing layer.
E&M
Evaluation & Management — CPT code category for clinician cognitive work (the most common physician billing codes).
ED
Emergency Department — the use case behind Augmedix Go ED (launched April 2024), purpose-built for high-acuity, multi-touch ED visits.
EHR
Electronic Health Record — Epic, Oracle Health, athenahealth, MEDITECH are the main targets for Augmedix write-back.
F1
F1 score — the harmonic mean of precision and recall. HCA pilot reported clinical documentation F1 > 0.8 on Augmedix output.
FHIR
Fast Healthcare Interoperability Resources — HL7’s modern API standard for healthcare data. Used in Augmedix’s EHR integrations and the Google Healthcare API.
FTC
Federal Trade Commission — the US consumer-protection regulator. No FTC actions documented against Augmedix.
GCP
Google Cloud Platform — the cloud substrate underneath every Augmedix component from STT through analytics.
GKE
Google Kubernetes Engine — the managed Kubernetes service hosting the encounter_service that subscribes to Pub/Sub.
GPO
Group Purchasing Organization — e.g. Vizient, which negotiates pricing on behalf of member health systems.
HCA
HCA Healthcare — the US hospital operator (1,400+ physicians piloting, 50+ hospitals). Augmedix’s exclusive ambient AI development partner post-Commure-acquisition.
HCC
Hierarchical Condition Categories — CMS risk-adjustment coding methodology supported by Augmedix Live/Assist coding workflows.
HDEv2
Healthcare Data Engine v2 — Google’s FHIR R4 harmonization product, in private preview as of June 2024. Aggregates multi-source healthcare data.
HIPAA
Health Insurance Portability and Accountability Act — US law governing PHI handling. Underlies every BAA in Augmedix’s architecture.
HITRUST
Health Information Trust Alliance — the framework. Augmedix holds HITRUST r2 (the rigorous risk-based, 2-year tier) since August 2023.
HL7
Health Level Seven — the standards organization and message format (HL7 v2) used by Augmedix to pull schedules from Epic (ADT/SIU feeds).
HPI
History of Present Illness — the narrative section of a clinical note describing the patient’s current complaint and its evolution.
ICD-10
International Classification of Diseases, 10th rev. — the diagnosis code set generated by Augmedix Go’s structured-data extraction.
IdP
Identity Provider — the SSO authority (often Imprivata in hospital settings) backing EHR-embedded SMART-on-FHIR launches.
IPO
Initial Public Offering — Augmedix uplisted to Nasdaq via reverse merger in December 2020 (ticker AUGX, $40M raised).
JSON
JavaScript Object Notation — the structured payload format Augmedix publishes to customer Pub/Sub topics (transcript, note, structured clinical data).
KPI
Key Performance Indicator — e.g. F1 score > 0.8 on HCA clinical documentation pilot, 99% patient consent acceptance in ED pilot.
LLM
Large Language Model — the model class behind MedLM, MedGemma, and Gemini in Augmedix’s multi-pass pipeline.
LOS
Length of Stay — an inpatient metric supported by Assist/Live coding workflows.
M&A
Mergers & Acquisitions — the Commure–Augmedix transaction (Oct 2024) is the relevant one here.
MDS
Medical Documentation Specialist — Augmedix’s job title for the trained remote scribe (often used interchangeably with “scribe” or RDS).
MedGemma
Google’s medical-domain Gemini-derivative LLM. Confirmed active in Augmedix’s Clinical Notes Engine pipeline.
MedLM
Google’s medical-tuned foundation-model suite (Med-PaLM 2 lineage). Augmedix was an early tester (Dec 2023); MedLM is fine-tuned on Augmedix’s 70K notes/week corpus.
MFA
Multi-Factor Authentication — Augmedix’s public product schedules do not specify MFA requirements; hospital IdPs typically enforce.
NLP
Natural Language Processing — the pipeline step between ASR and LLM generation. Augmedix runs proprietary NLP for entity extraction and section assignment.
OCR
Office for Civil Rights (HHS) — the HIPAA enforcement body. No OCR actions documented against Augmedix.
PE
Physical Exam — the section of a clinical note recording the clinician’s physical examination findings.
PHI
Protected Health Information — the HIPAA-defined identifiable patient data category. Every Augmedix BAA exists to govern PHI flow.
Pub/Sub
Google Cloud Pub/Sub — asynchronous messaging service used to publish structured encounter JSON to customer-owned topics.
QA
Quality Assurance — the human review step inside NoteBuilder where MDSs polish AI drafts and the discrepancies become fine-tune training signal.
RCM
Revenue Cycle Management — the billing and reimbursement workflow Commure layers on top of Augmedix documentation via Athelas.
RDS
Remote Documentation Specialist — alternate term for MDS, used in some Augmedix materials. ~70% of RDSs were resident in India at one point.
ROS
Review of Systems — the section of a clinical note inventorying patient symptoms by organ system.
SLA
Service Level Agreement — Augmedix commits 99.0% uptime on its Pub/Sub enterprise-data-export service.
SMART on FHIR
The OAuth-based EHR-embedded app launch standard used when the Augmedix app launches inside Epic Hyperspace/Hyperdrive.
SOC 2
System and Organization Controls 2 — AICPA assurance framework. Not publicly confirmed for Augmedix; HITRUST r2 is the primary third-party posture.
SPAC
Special Purpose Acquisition Company — the vehicle (Malo Holdings) used in Augmedix’s 2020 reverse-merger IPO.
SSO
Single Sign-On — the hospital IdP authentication pattern (often Imprivata-backed) inherited by Augmedix in SMART-on-FHIR EHR-embedded launches.
STT
Speech-to-Text — Google Cloud’s diarized ASR service. Synonym for ASR in this context.
VPC SC
VPC Service Controls — GCP perimeter-isolation feature commonly used to wrap Vertex AI / Healthcare API in healthcare deployments. Usage at Augmedix unconfirmed.
VWAP
Volume-Weighted Average Price — the 30-day VWAP baseline against which Commure’s 169% acquisition premium was measured.
YoY
Year-over-Year — the growth metric used in Augmedix financial reporting (e.g. +45% YoY revenue growth in 2023, +44% YoY clinician growth).
Sources
Synthesized from publicly available sources, prioritized for technical and architectural detail:
- Google Developer Community — “The future of clinical data: building an intelligence engine with Augmedix & Google Cloud” (June 2024). The richest single technical source — pipeline diagram, GCP service inventory, retention figures.
- Augmedix S-1 IPO Prospectus (SEC EDGAR, 2020) — corporate structure, customer concentration, NoteBuilder framing.
- Augmedix Form 10-K FY2023 and Q4 2023 earnings release (GlobeNewswire) — financial scale, clinicians-in-service, ARR.
- Augmedix Go and Live product schedules (augmedix.com) — the closest public document to a technical spec.
- Augmedix Privacy Policy (augmedix.com/privacy-policy/current) — subprocessors, data retention general terms.
- HITRUST r2 announcement (GlobeNewswire, August 2023) — certification scope.
- Commure acquisition press releases (July–October 2024) and Hospitalogy analysis (October 2025) — deal terms, integration thesis.
- HCA partnership (GlobeNewswire, April 2023) and Hospitalogy interview with Ian Shakil — ED pilot details, the “conversational graph” framing.
- MedLM launch announcement (Google Cloud Blog and Fierce Healthcare, December 2023) — Augmedix as early tester.
- Oracle Validated Integration announcement (GlobeNewswire, December 2024); MEDITECH Expanse Now embed (Commure press release, September 2025); Vizient contract (Commure press release, January 2025).
- TBS News Bangladesh — on the Dhaka scribe workforce.
- Augmedix Epic Showroom listing (id=1103) and Oracle Marketplace listing — marketplace presence.
Built for GG Prompts using research from public sources. Augmedix is a real company (now part of Commure, marketed as “Commure Ambient”); this map summarizes publicly available information as of May 2026. Items marked [unconfirmed] are inferred from typical GCP-Healthcare patterns and labelled explicitly. Not affiliated with Augmedix, Commure, Google Cloud, or any health system named on this page.