A Documentation Architecture Built for Meditech Expanse Hospitalists
Hospitalist documentation inside Meditech Expanse rarely fails because clinicians lack diligence. It fails because the inpatient day is a series of interruptions: a rapid admission, three progress notes before rounds end, a consult request, a cross-cover handoff, and a discharge summary that must be reconciled against orders, medications, and disposition. Every one of those note types lives in a different Expanse workflow, and every context switch between ambient capture and manual charting compounds cognitive load. Merry AI was designed to remove that switch entirely by writing structured drafts into the exact note field you already have open.
Before we describe the mechanism, a precise statement of scope is warranted. MEDITECH publishes RESTful integration infrastructure and Expanse FHIR APIs, and vendors such as Suki have achieved certified ambient integration directly through the Expanse Document APIs. Merry AI takes a deliberately different architectural path. We do not assert a MEDITECH-certified Document API write-back, a registered OAuth service account, or automated routing into a named provider queue. Instead, Merry AI injects a finalized draft into the rendered DOM note field of the encounter you personally opened, leaving the signature and queue-routing entirely inside Expanse where they belong.
This distinction matters for both compliance accuracy and clinician trust. Claiming native API write access to a closed configuration you have not certified is a defect, not a feature. The honest, defensible description is that Merry AI generates AI-assisted documentation and places it into the field your keyboard already targets, inheriting your authenticated session rather than creating a parallel privileged pathway into the chart.
How Browser-Native DOM Injection Actually Works
When you open an active inpatient encounter in Expanse and begin an admission or progress note, the Merry AI Chrome Extension is already running inside that same browser tab. It listens to the ambient encounter, produces a structured draft using the appropriate hospitalist template, and identifies the note text-entry element by its stable DOM selector. On your explicit command, it writes the draft into that element exactly as typed characters would appear. There is no intermediate server posting into your chart and no background process acting on your behalf.
Because the injection travels through the same DOM your cursor uses, it operates entirely within your existing session context, session timeout, and Expanse audit attribution. Merry AI never selects the patient, never advances the note state, and never signs. The clinician remains the author of record, the reviewer, and the signer. This is the architectural inverse of an API robot: the human is in the loop by design, not by policy overlay.
Encounter Matching and the Hard Stop
Patient and encounter fidelity is non-negotiable on an inpatient service where a single provider may touch twenty charts before noon. The extension reads the patient identifier, encounter ID, facility, and attending already displayed in your Expanse window and presents them in a confirmation banner. If that context is missing, ambiguous, or changes while the note is open, the extension refuses to inject and issues a hard stop. Combined with your own final signature, this produces two independent checks binding the note to the correct admission, transfer, or observation encounter.
Note Typing and Provider Queue Behavior
Merry AI produces drafts, not finalized documents. The generated content lands in the note field as an unsigned draft; the transition to a signed, queue-routed, or co-signature-pending state occurs only when you act inside Expanse. This respects hospital-specific co-signature rules for residents and advanced practice providers, preserves the distinction between an amendment and an overwrite, and ensures an automatically drafted discharge narrative is never mistaken for a completed, signed discharge summary.
Integration Architecture Compared
The table below contrasts the three practical paths a hospitalist group can take to get ambient AI output into Expanse. Each trades differently on deployment overhead, speed, and cost.
| Architecture | Deployment Overhead | Latency | Annual Fee | Clinician as Author |
|---|---|---|---|---|
| Traditional API Marketplace Integration | High: vendor certification, OAuth scopes, endpoint registration, upgrade re-testing | API round-trip plus queue processing | Marketplace and per-transaction fees | Depends on write-back config |
| Manual Copy-Paste | Low: none, but fully manual | Human transcription time; context switching | None | Yes, but error-prone |
| Merry AI DOM Injection | Minimal: per-clinician Chrome Extension, no server or firewall change | Sub-second field mapping (0.4s) | Flat practice subscription, no marketplace fees | Yes, always |
The marketplace path is the correct choice when a health system requires certified, autonomous write-back at enterprise scale and can absorb the certification timeline and fees. Manual copy-paste is free but reintroduces exactly the transcription burden ambient AI exists to eliminate. DOM injection occupies the pragmatic middle: near-zero deployment, sub-second placement, and unambiguous clinician authorship.
Clinical Safeguards for the Inpatient Setting
Hospitalist workflows carry documentation risks that a responsible integration must address explicitly rather than assume away. The Association of Medical Directors of Information Systems consensus on inpatient EHR documentation warned that copy-forward and unverified auto-population degrade the integrity of the legal record. Merry AI answers these concerns with visible AI provenance, template-enforced section completeness, and warnings for missing high-acuity fields such as code status, allergies, and medication reconciliation.
High-acuity data is never silently inferred. Medication changes, critical results, and diagnoses surface as items requiring clinician confirmation, and generated content stays visually distinct from your edits so provenance remains auditable at signature. Because Merry AI cannot sign, cannot route, and cannot select the encounter, it structurally cannot substitute for the required handoff, co-signature, or discharge workflows your facility mandates.
For groups documenting longitudinal outpatient panels alongside inpatient service, Merry AI structures E/M narrative to support appropriate code selection, including the G2211 complexity add-on, consistent with CMS guidance and the AMA CPT framework. The system supplies the supporting language; the coding decision remains the clinician's, preserving the separation between documentation assistance and billing judgment.
Deploying Merry AI Across Your Hospitalist Group
Getting started is deliberately unhurried and low-risk. A clinician installs the Chrome Extension into their browser profile, selects or customizes the relevant inpatient templates from the Template Center at Access 1,000+ EHR Clinical Templates, and begins drafting inside their existing Expanse session. There is no server provisioning, no OAuth negotiation with MEDITECH, and no firewall change to schedule with IT.
Under an executed Business Associate Agreement, Merry AI retains zero audio and zero transcript, processes over TLS to a SOC 2 environment, and provides a data-flow diagram and subprocessor list to support your security review. This posture keeps the integration aligned with HIPAA minimum-necessary principles while avoiding the standing privileged credentials a marketplace integration would require.
To evaluate the fit for your service line, the most efficient next step is a structured review of your current admission-to-discharge documentation path. Book a Schedule a 15-Minute Workflow Audit and we will map your Expanse note types to Merry AI templates on a live call. When you are ready to roll out across the group, Explore Merry AI Practice Partner Plans for flat, predictable pricing with no per-transaction marketplace fees.
