Case Study
How a Clinical Team Halved the Time It Takes to Review a Patient Record.
A clinical intelligence layer for EMRs, designed to bring relevant history, recorded conditions, and clinical findings into one review.
- Record review, down from 20 minutes
- 10 Min
- Clinical summary preparation, down from 15 minutes
- 5 Min
- Saved per record across review and summary preparation
- 20 Min
The Challenge
The Record Holds the Information. The Review Needs the Context.
Bring the evidence together so the clinician can assess it in context.
The Business Context
The work brought patient-related information and supporting data into a healthcare intelligence layer for analysis of medical conditions and diagnostic context.
The focus was making healthcare information usable for analysis. Patient data supplies the clinical context; language models need that context to contribute to the work. The platform brings those elements together, with EMR use as the direction for the experience.
- Which entries are relevant to the condition being reviewed?
- How have the recorded findings changed across encounters?
- What source information supports an answer, and what remains uncertain?
What We Built
We built the healthcare data intelligence layer around the genius office Data Intelligence Platform, connected with Vercel eve, Claude API, and pretrained models sourced through Hugging Face. The work combines patient-related and supporting information for analysis. Condition-focused chart views and source-linked questions are potential EMR applications of that foundation.
The Solution
Three Questions a Clinician Asks, and One Boundary to Keep.
01 / 04
Every entry in the record
- Clinic note · 12 Jan
- Lab result · 03 Feb
- Administrative letter · 09 Feb
- Imaging report · 20 Feb
The Need
Which Entries Matter for the Condition Being Reviewed?
The record holds the information, but the entries that matter for one condition are spread among everything else.
Entries related to the condition in view
- Clinic note · 12 Jan
- Lab result · 03 Feb
- Administrative letter · 09 Feb
- Imaging report · 20 Feb
What the Layer Makes Possible
A View of the Record by Condition
A clinician could choose a condition and review the related history, recorded diagnoses, notes, and findings together, each with its source and date.
How It Works
The Information Brought TogetherData Intelligence PlatformPatient-related and supporting information in one analysis layer
Pretrained ModelsHugging FaceSelected models support the analysis of the record
- 03
Source and Date KeptEvery item shown keeps where and when it came from
02 / 04
Findings sit in separate entries The Need
How Have the Findings Changed across Encounters?
Seeing change means rebuilding the sequence by hand across separate entries.
- 12 Jan
- 03 Feb
- 20 Feb
- 15 Mar
The same findings, in date order What the Layer Makes Possible
Findings in Date Order
A timeline could bring relevant findings into chronological order, so a clinician can inspect the changes directly.
How It Works
- 01
Consistent Dates and UnitsFindings are aligned before they are placed in order
- 02
Each Finding Linked to Its SourceThe clinician can open the entry behind any point
- 03
For Inspection, Not DiagnosisA trend supports review. It does not by itself establish a diagnosis
03 / 04
What in the chart supports this recorded diagnosis?
An answer with nothing to check it against
The Need
What Supports an Answer, and What Remains Uncertain?
An answer is only useful to a clinician if they can see what it rests on, and what is missing.
What in the chart supports this recorded diagnosis?
A response tied to the supporting entries
Clinic note · 12 JanLab result · 03 FebNo imaging on record
What the Layer Makes Possible
Answers Tied to the Record
A clinician could ask about recorded conditions or findings and receive a response tied to the supporting entries. Missing and conflicting evidence stays visible.
How It Works
Coordinating the AnalysisVercel eveThe agent layer gathers the permitted context for the question
A Response with Its SourcesClaude APILanguage models contribute once they have the clinical context
- 03
Gaps Stay VisibleMissing or conflicting information is shown for review
04 / 04
Patient record
- Name
- Identifier
- Conditions
- Findings
AI model
What reaches the model has to be controlled The Need
Patient Identity Has to Stay Out of the Models
Using AI on healthcare information only works if what reaches the models is controlled.
Patient record
- Nameheld back
- Identifierheld back
- Conditions
- Findings
AI model
Only the permitted context reaches the model What the Layer Makes Possible
A Privacy Boundary before Any Analysis
Identifying information was kept out of model inputs, and analysis uses only the information permitted for the task.
How It Works
- 01
Identifying Details Held BackThey are not passed to the models, as confirmed by the project owner
Covered UseClaude APIUnder a signed HIPAA Business Associate Agreement and zero data retention
- 03
Only What the Task NeedsAnalysis uses the information permitted for that task
These arrangements apply to the covered Claude API use. They are not a certification of the whole platform.
From the Patient Record to Evidence for Review
The Information to Connect
- EMR and Supporting Data Sources
Patient History
Recorded Conditions
Clinical Findings
Patient-related and supporting information supplies the context for analysis. The connection to an individual EMR depends on its data access and integration capabilities.
Proposed EMR Analysis Workflow
- 01
Privacy Boundary
Prepare the Context
Select relevant data and apply the agreed privacy boundary
02 · The Platform
Connect the Information
Bring the analysis context into the genius office platform
03 · AI
Prepare the Analysis
Use Vercel eve, Claude API, and selected Hugging Face models
04 · Clinician
Review in Context
Present findings and source references for clinician review
The proposed output supports clinical review. Diagnosis and treatment decisions remain with qualified clinicians.
Technologies Used
Data Intelligence Platform
Built by genius office
Vercel eve
Agent Layer for Analysis
Hugging Face
Pretrained Models for Analysis
Claude API
Enterprise AI · HIPAA BAA & Zero Data Retention
Potential Applications in the EMR
- 01
Build a Condition-Based View of the Record
A clinician could choose a condition and review related history, recorded diagnoses, notes, and findings together. The view would retain the source and date of each item so its context stays visible.
Technical Detail: Build a Condition-Based View of the Record
A condition-focused view needs a consistent way to associate entries with the condition and preserve the original records. The connection and supported data types depend on the EMR; no particular interoperability standard is claimed here.
- 02
Keep Identifying Information Outside Model Inputs
Identifying information was kept out of model inputs. The Claude API integration used a signed HIPAA Business Associate Agreement (BAA) and zero data retention, as confirmed by the project owner. Analysis uses the information permitted for the task.
Technical Detail: Keep Identifying Information Outside Model Inputs
The owner confirmed the HIPAA BAA and zero data retention arrangements for Claude API. Their coverage is specific to the applicable agreement, enabled configuration, and eligible API features; it does not automatically extend to Vercel eve, Hugging Face models, or the full platform. Excluding identifying information also requires attention to free-text content, not only structured fields.
- 03
Ask Questions with the Record in View
The proposed experience would let a clinician ask about recorded conditions or findings and receive a response tied to the supporting entries. Missing evidence and conflicting information would remain visible for review.
Technical Detail: Ask Questions with the Record in View
Illustrative question: “What evidence in the chart supports this recorded diagnosis?” The platform would supply the permitted context, with Vercel eve coordinating analysis and Claude API and selected Hugging Face models supporting language tasks. This is an example application, not a claim of measured diagnostic accuracy.
- 04
Show Changes Across Encounters
A timeline could bring relevant findings into chronological order, helping a clinician inspect changes without reconstructing the sequence across separate entries.
Technical Detail: Show Changes Across Encounters
A useful timeline needs consistent dates and units, with each displayed finding linked to its source. Trend presentation would support inspection of the record; it would not by itself establish a diagnosis.
The Results
More Time for Clinical Review. Less Time Preparing the Record.
Record Review and Summary Preparation
- Before the Analysis Layer
- 35 Min
- With the Analysis Layer
- 15 Min
20 Minutes Saved per Record
Record review fell from 20 to 10 minutes, and clinical summary preparation from 15 to 5 minutes. Together, the two tasks took 15 minutes instead of 35, saving 20 minutes per record.
A Connected Analysis Layer
The delivered layer brings healthcare information and pretrained-model capabilities together through the genius office Data Intelligence Platform and Vercel eve, with Claude API and models sourced through Hugging Face. It provides a foundation for examining medical conditions and diagnostic context using the supplied data.
Less Time Preparing Each Record
The project owner confirmed shorter review and summary preparation times after checking project correspondence. The combined saving gives the team more time for the work around each record, while clinicians remain responsible for reviewing the information and making clinical decisions.
Time figures were confirmed by the project owner after checking project correspondence. The combined saving is calculated as (20 + 15) − (10 + 5) = 20 minutes per record. The measurement period and sample size were not supplied. These workflow results do not establish diagnostic accuracy or improved patient outcomes.
Where the Platform Can Help
- Clinicians
- Review relevant information and its source context within the patient-record workflow.
- EMR Product Teams
- Explore how a clinical intelligence capability could fit into their existing software and user experience.
- Healthcare Data Teams
- Define the data made available for analysis and validate how it is represented in the output.
Principles for EMR Applications
Keep the Clinical Question in Focus
Define the question the application should support and the information needed to answer it. A broad model capability is only one part of that workflow.
Preserve the Source Context
Clinical information needs its original meaning, date, and source. A summary should help the reader return to the underlying record.
Make the Data Boundary Explicit
Determine which information can be used for analysis and how identifying details are handled throughout the integration.
Evaluate the Intended Use
Assess an EMR application against its actual review task. Model availability alone does not demonstrate clinical usefulness or accuracy.