Ambient documentation
Real-time audio transcription via WhisperX, Parakeet, and Canary ASR engines, routed through a dedicated mTLS gateway.
Artificial Intelligence
AI integration that allows clinicians to focus on patients,
not charting.
Ambient documentation, clinical extraction, and SOAP structuring - running in the same workspace as the chart.
Encounter Intelligence Layer
From audio to chart, with the physician in the loop.
Chained pipeline, per-domain confidence gates, conflict-safe physician review.
Speaker attribution
Clinical vocabulary density and turn-order heuristics identify physician versus patient speech without recording assumptions.
Clinical extraction
Eight extraction types - problems, assessments, medications, lab orders, imaging orders, referrals, history, follow-ups - using local large language models.
Chart reconciliation
ICD-10 exact match, embedding cosine similarity via nomic-embed-text, and Levenshtein fallback to map extracted concepts to existing chart elements.
Physician review
A two-column review interface with confidence badges, transcript drawer, and conflict-safe bulk approval. Physician stays in control.
Chart write services
Problem, order, referral, history, and follow-up writers, each with applied-payload snapshots and a reversal window.
Streaming dictation
Live transcription, SOAP structuring on commit.
250 ms WebSocket chunks. Auto-segmented into S, O, A, P. Physician approves before any chart write.
- WhisperLiveKit WebSocket streaming with 250-millisecond chunks and voice activity detection
- Parakeet as the primary English engine; Canary as fallback
- SOAP LLM post-processing via llama3.3:70b on a local DGX Spark - physician reviews before commit
- Multilingual transcription across 25 European languages with consent-flag enforcement
End Visit to Sign
One press of End Visit. One atomic Sign.
The full ambient pipeline runs in parallel the moment the visit ends. The physician reviews everything in a single overlay, then signs in one transaction with per-section savepoint rollback.
End Visit auto-trigger
Pressing End Visit launches a Celery chord that runs SOAP structuring, the billing copilot (ICD-10 / E&M / CPT), the order-suggest agent, and a prior-auth packet builder in parallel. Server-sent events notify the chart as each stage completes.
Six-tab review overlay
EncounterReviewOverlay sits between End Visit and Sign with Note, Problems, Orders, Billing, Follow-up, and Compliance tabs. Confidence-gated AI suggestions surface inline with accept, reject, and edit controls per card.
Atomic finalize-and-sign
POST /encounters/{id}/finalize-and-sign/ commits problems, orders, billing, follow-up, and compliance attestations inside a single transaction with per-section savepoints. Any failed section rolls back without losing the others.
Co-sign and EPCS
Trainee notes route to an attending co-sign queue. Controlled-substance orders surface the EPCS modal for DEA two-factor attestation before the prescription is included in the signed commit.
AI integrations console
Cloud AI with cost discipline and BAA gating.
Every AI call site is registered, every override is reviewable, every dollar is tracked, and every cloud provider runs behind a HIPAA BAA gate.
- Read-only inventory of every AI call site with model, host, and last-run metadata
- Editable override drawer with server and model dropdowns, replacing free-text configuration
- Six cloud provider adapters - Anthropic, OpenAI, xAI, Gemini, Azure OpenAI, AWS Bedrock - with HIPAA BAA gating
- Per-integration prompt overrides with an approval queue
- Daily spending caps enforced as an HTTP 429 ceiling before the provider call leaves the building
- Pricing table populates cost in USD on every AI event; health dashboard shows 24-hour and 30-day spend with cap-warning badges
- log_ai_event slug tagging, Prometheus metrics, and health probes across every adapter
Coding and billing AI
Coding suggestions from the note.
ICD-10 and CPT codes derived from encounter context, integrated directly into the clinical workspace.
- ICD-10 and CPT suggestions from encounter context
- AI claims review and denial risk prediction
- AI medication safety checking
- AI quality narrative builder
- AI pre-visit briefing
Governance
Phase-gated, audited, and PHI-safe.
Read-only, low-risk write, and moderate-risk write phases. Every action in the execution ledger.
- Phase-gated activation across read-only, low-risk write, and moderate-risk write phases
- Per-domain confidence gates
- PHI-safe log assertions and automated redaction checks
- Consent enforcement middleware on every data domain endpoint
- AI execution ledger for cross-tool continuity and audit
- AI governance board with task approval workflows
Our AI stack
Integrated AI Models.
Speech, embedding, and language models, with explicit named placement.
| Layer | Engine |
|---|---|
| Speech (English) | Parakeet, WhisperX |
| Speech (multilingual) | Canary (25 European languages) |
| Streaming transport | WhisperLiveKit, WebSocket, 250 ms chunks |
| SOAP structuring | llama3.3:70b, llama3-med42v2 |
| Clinical embeddings | nomic-embed-text |
| Routing | Dedicated mTLS gateway, Nginx-based Docker container |
Next
See how the platform handles security and HIPAA.
Field-level PHI encryption, audit logging, the resolved findings of an externally audited security review, and an honest list of where certifications stand today.