The Problem
Unstructured conversation. Structured records required.
A clinical consultation is a natural conversation — but the EHR requires structured, coded data: ICD-10 diagnoses, RxNorm medication names, CPT procedure codes, standardized vital sign fields. Translating from one to the other manually takes 10–20 minutes per visit.
ELLEXMED uses large language model NLP (Gemini 2.5 Pro) to perform this extraction automatically — identifying the clinical entities in the transcript and mapping them to the correct structured fields in a single inference pass.
The Design Principle
AI is the author. You are the editor.
No auto-filled field is ever saved directly to the patient record without explicit clinician review. The EHR form opens with AI-suggested content — the clinician reads, edits, confirms, and submits. This is not autonomous documentation.
This design matters for liability, compliance, and clinical accuracy. The AI removes the burden of first-draft creation. The clinician retains full responsibility for the final record.
Fields extracted and filled automatically
Chief Complaint & HPI
Extracts the patient's primary complaint and history of present illness from consultation narrative
ICD-10 Diagnoses
Maps diagnostic statements to ICD-10-CM codes — presented as suggestions for clinician confirmation
CPT Procedure Codes
Identifies procedures, investigations, and interventions mentioned and suggests CPT billing codes
Medications & Allergies
Extracts current medications, new prescriptions, dose changes, and allergies with reactions stated
Vitals & Exam Findings
Captures blood pressure, heart rate, weight, temperature, and physical examination observations
Assessment & Plan
Structures the clinical impression and management plan from the transcript into SOAP-format fields
FAQ
Automated EHR filling — answered.
Which EHR fields does ELLEXMED automatically fill from a transcript?
ELLEXMED extracts and fills the following structured fields: Chief Complaint, History of Present Illness (HPI), Past Medical History (PMH), medications (current and newly prescribed), allergies mentioned, physical examination findings, vital signs, ICD-10 diagnoses, CPT procedure codes, assessment (clinical impression), and plan (next steps including investigations and follow-up). All fields are presented as drafts for clinician review and editing.
How does ELLEXMED extract ICD-10 codes from a clinical transcript?
ELLEXMED uses Google Gemini 2.5 Pro to perform NLP entity extraction on the visit transcript. The model identifies diagnostic statements (e.g., 'the patient has type 2 diabetes mellitus with hypertension'), maps them to ICD-10-CM codes, and presents them as suggested diagnoses. The clinician reviews and confirms or modifies each code before the visit is finalized.
Can doctors edit the AI-generated EHR content?
Yes — all auto-filled content is editable before submission. ELLEXMED is designed around the principle that AI is the author of a first draft and the clinician is always the editor and final signatory. No AI output is ever saved directly to the patient record without explicit clinician review and sign-off.
Does ELLEXMED integrate with Epic, Cerner, or other EHR systems?
ELLEXMED is a standalone clinical platform with its own built-in EHR module rather than a plugin for third-party EHR systems. It manages its own patient records, visit documentation, and clinical data. Enterprises requiring custom integrations with external EHR systems should contact the sales team.
How accurate is the ICD-10 and CPT extraction?
Accuracy depends on transcript quality and the specificity of clinical language used. The AI performs best when the transcript contains explicit diagnostic statements and procedure names. Complex cases with ambiguous language will produce lower-confidence suggestions, which are flagged for clinician review. ELLEXMED does not guarantee coding accuracy for billing purposes — final code selection is the clinician's responsibility.
What AI model powers the EHR auto-fill feature?
EHR auto-fill is powered by Google Gemini 2.5 Pro via the Genkit framework. This model was selected for its strong performance on structured extraction tasks from medical text, including ICD coding, medication identification, and clinical note structuring.