Page 163 - AI for Good-Innovate for Impact Final Report 2024

AI for Good-Innovate for Impact

Use case – 37: Multilingual Medical Language Models: A Path to

Improving Lay Health Worker Effectiveness 37-Ethriva

Country: United States

Organization: Ethriva

Contact Person: Agasthya Gangavarapu, august@ ethriva .a,

Ananya Gangavarapu, ananya@ ethriva .com

37�1� Use case summary table

Domain Healthcare

Problem to be Access to medical care, inefficient management of waiting times, long
addressed queues.

Key aspects of Local language chatbot, medical records, pre-health assessment, diag-
the solution nostics, summary report generation, doctor dashboard, mobile based
solution.
Technology Large Language Model (LLM), Machine Translator (MT), API
keywords

Data availability Private
Metadata (type Medical dialog data, adverse events, clinical notes, medical records
of data)
Model Training Fine-tuned Llama 3 70B model with GQA and quantization. Seamless
and fine tuning M4T-Large is finetuned with context and cultural information. NVIDIA
NeMo Guardrails is customized to filter out the prompts that are toxic.

Testbeds or https:// arxiv .org/ abs/ 2404 .08705
pilot deploy-
ments

37�2� Use case description

37�2�1� Description

This innovative project introduces a transformative use case targeting the enhancement of
healthcare delivery by empowering Community Health Workers (CHWs) in Low- and Middle-
Income Countries (LMICs) through the integration of Large Language Models (LLMs) with
machine translation technologies. Aimed at addressing the critical shortfall of healthcare
workers, the initiative seeks to navigate the complex challenges of linguistic barriers, cultural
nuances, and the scarcity of tailored medical dialog datasets, which collectively impede the
efficacy of CHWs in remote and underserved regions.

147