Loading…
Hire an LLM Engineer, Nashville
This is for the VP of Health Informatics or CMIO at HCA Healthcare (Nashville headquarters), Vanderbilt University Medical Center, or Ardent Health Services. The problem you are solving: physician documentation burden. Doctors spend 2 to 3 hours per day on notes, 70% of which is templated content that could be auto-populated from existing data.
An ambient documentation LLM drafts the SOAP note from transcribed encounter audio or post-encounter dictation, pulls relevant history from the EHR, and pre-populates required fields. The physician reviews and signs, they do not write from scratch. The engineering requirements are specific: accurate ICD-10 and CPT code suggestions, HIPAA compliance via a signed Business Associate Agreement, and Epic EHR integration via SMART on FHIR.
Tell us your EHR platform, target specialty, and your documentation burden metric.
At HCA Healthcare's 180+ hospitals, physician burnout is partly a documentation problem. A 2023 AMA study found that for every hour of patient care, physicians spend nearly 2 hours on EHR documentation. For a primary care physician seeing 20 patients per day, that is 6 to 8 hours of documentation work, most of which is templated note structure rather than clinical judgment.
The technical problem: an ambient documentation LLM needs to produce a note that accurately reflects what happened in the encounter, uses the correct SOAP structure for the specialty, suggests appropriate ICD-10 codes, and integrates smoothly into the physician's existing Epic workflow. The physician should spend 5 to 10 minutes reviewing and signing, not 20 minutes correcting.
Audio transcription is typically handled by a dedicated transcription service (Deepgram, AssemblyAI, or AWS Transcribe Medical) before the LLM step. The LLM receives the transcript, the patient's problem list and current medications from the Epic FHIR API, and a specialty-specific SOAP template. It generates the structured note draft.
ICD-10 suggestion runs as a second pass on the Assessment section. CPT code suggestion for procedure notes uses a separate prompt calibrated for your payer mix and specialty billing patterns. At Vanderbilt UMC, the academic medical center context requires billing accuracy that a community hospital may be less sensitive to.
Receives post-encounter transcript from your transcription service. Generates specialty-calibrated SOAP draft using patient context from Epic FHIR.
Pulls active problem list, current medications, relevant lab results, and last encounter summary via Epic SMART on FHIR API before generation.
Second-pass analysis of the Assessment section produces ranked ICD-10 suggestions with clinical rationale. Physician selects, not auto-submits.
Draft writes to Epic note composer via the Notes API. Physician reviews and signs within the standard Epic workflow, not a separate application.
BAA with model provider, PHI data minimization in prompts, audit logging of all LLM calls, encryption in transit and at rest. Documentation included.
Separate SOAP templates for each service line: primary care, orthopedics, oncology, cardiology. Each calibrated with 5-10 sample notes from that specialty.
The discovery call includes your Epic Technical Architect or EHR integration lead, your privacy officer, and a physician champion from the target specialty. We review: your Epic FHIR API access configuration, your transcription vendor, the SOAP note format for the target specialty, and your ICD-10 coding accuracy requirements.
We build for one specialty first and run a pilot with 5 to 10 physicians before expanding. The pilot measures three things: time to review and sign the draft (target under 8 minutes), edit rate per note section (baseline for improvement), and ICD-10 suggestion acceptance rate (target above 85% for primary care).
Build time: 10 to 16 weeks for a single specialty with Epic integration, HIPAA compliance documentation, and a pilot cohort evaluation.
Tell us your EHR platform, target specialty, current documentation time per physician, and your privacy requirements. We will reply with a technical approach and rough scope within two business days.