Mobile Medical Note Generation: Local Transcription, Cloud LLM Summarization

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing medical note generation systems are costly due to cloud-based transcription, lack data privacy, and are not HIPAA compliant, with ongoing expenses and security risks.

Innovation Solution

A local speech recognition model running on a mobile device for real-time transcription combined with a cloud-based large language model for summarization, ensuring HIPAA compliance and reducing costs by eliminating cloud-based processing.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If cloud-based transcription is used, then transcription accuracy is improved, but ongoing costs and data security risks increase

Engineering Contradiction:
Improvetranscription accuracyVSAvoiddata security risks
Core Design Contradiction:
Measurement precisionVSObject-affected harmful factors

Solution Approach 1:

The patent extracts the transcription processing function from cloud-based systems and implements it locally on mobile devices using on-device machine learning models. This extraction eliminates the need to transmit sensitive medical data to cloud servers, thereby maintaining transcription accuracy through sophisticated local models while eliminating data security risks associated with cloud transmission and storage.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent introduces an on-device transcription model as an intermediary between the audio input and the medical note generation system. This local model processes speech data directly on the mobile device, serving as a secure intermediary that prevents sensitive data from leaving the device while still providing accurate transcription for subsequent medical note generation.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Measurement precision

If cloud-based transcription is used, then transcription accuracy is improved, but ongoing operational costs increase

Engineering Contradiction:
Improvetranscription accuracyVSAvoidongoing operational costs
Core Design Contradiction:
Measurement precisionVSLoss of energy

Solution Approach 1:

The patent extracts the transcription processing function from cloud-based systems and implements it locally on mobile devices using on-device machine learning models. This extraction eliminates the need to transmit sensitive medical data to cloud servers, thereby maintaining transcription accuracy through sophisticated local models while eliminating data security risks associated with cloud transmission and storage.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent enables the mobile device to perform transcription processing independently through on-device machine learning models. This self-service capability allows the device to transcribe medical conversations locally without requiring ongoing cloud service subscriptions, thereby eliminating recurring operational costs while maintaining high transcription accuracy.

Inventive Principle:
Principle #25Self-service

3Reliability

If local speech recognition is implemented, then data privacy is enhanced, but model accuracy and adaptability may be reduced

Engineering Contradiction:
Improvedata privacyVSAvoidtranscription accuracy
Core Design Contradiction:
ReliabilityVSMeasurement precision

Solution Approach 1:

The patent employs parameter changes in the form of on-device machine learning models that are specifically optimized for local execution. These models use quantization and other optimization techniques to reduce computational requirements while maintaining high transcription accuracy, thereby achieving both data privacy through local processing and high accuracy through optimized model parameters.

Inventive Principle:
Principle #35Parameter changes

4Power

If cloud-based processing is used, then computational power is sufficient, but HIPAA compliance and data security are compromised

Engineering Contradiction:
Improvecomputational powerVSAvoidHIPAA compliance
Core Design Contradiction:
PowerVSReliability

Solution Approach 1:

The patent extracts the transcription processing function from cloud-based systems and implements it locally on mobile devices using on-device machine learning models. This extraction eliminates the need to transmit sensitive medical data to cloud servers, thereby maintaining transcription accuracy through sophisticated local models while eliminating data security risks associated with cloud transmission and storage.

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentUS20250336393A1System and method for mobile device based real-time medical note generation utilizing a local speech recognition model and cloud based large language model
Publication Date: 2025.10.30 ZHOU LING

AI summary

The present disclosure relates to a system and method for generating medical notes in real-time using a mobile device. The system employs a local, real-time speech recognition model to transcribe medical conversations accurately without relying on cloud-based processing. The transcribed text is then sent to a cloud-based LLM for summarization and generation of a comprehensive medical note. The local speech recognition model is pre-trained on a dataset comprising numerous medical terms and phrases, and it can be iteratively improved through user feedback and additional training. This innovative approach eliminates the ongoing costs associated with cloud-based transcription services while ensuring data privacy and security by keeping sensitive patient information on the mobile device. The system offers the ability to deliver a better product at a price point an entire order of magnitude lower than our competitors.