Exploring Generative-AI Models for Medical Coding Technology
Project Goals
Since the release of generative-AI models, companies in many industries are looking to leverage this revolutionary technology. Soon after the release of the first generation large language models (LLMs), NotoVox wanted to survey the leading gen-AI platforms, and explore the feasibility of custom training of LLMs with medical coding data as a new framework for their service. Project details are protected under an NDA, so this discussion is generalized and images are open-source representations.
Project Results
The primary deliverable was a detailed report of findings from this Gen-AI investigation provided to the CTO. The work began with background research on how LLMs compare to traditional AI and overviews of the leading Gen-AI cloud platforms (e.g. Microsoft Azura OpenAI, Amazon Bedrock and Titan Models, and Google PaLM, Vertex AI, and MakerSuite). Next, I documented core concepts in prompt design and LLM model settings, such as setting the “temperature” parameter to minimize randomness and produce more prescriptive results. Finally, prototyping on the Google Vertex AI and MakerSuite platform was conducted with a sample set of medical codes (SNOmed and ICD10). This work involved learning about data import formats, cleansing the set of training data containing medical symptom-to-code pairs, setting up training jobs, and executing a series of challenging queries to test the network's ability to distinguish between conditions with similar symptoms. Overall results were mixed as these were the initial releases of the Gen-AI platforms. However, rapid progress was being made and we could see that new versions were yielding better results.
Skills Used
Secondary market research on AWS Marketplace as a SaaS platform for medical applications
Project Goals
NotoVox wanted to investigate Amazon's cloud offerings and AWS Marketplace as a SaaS platform for the next generation NotoVox architecture. The evaluation included an in-depth exploration of the benefits and costs of hosting an application on AWS and promoting in the AWS Marketplace
Project Results
The primary deliverable was a detailed report outlining the benefits, categories, pricing models and ratings methodology within AWS Marketplace. AWS technologies for SaaS offerings were also explored, including: Elastic Compute Cloud (EC2), Simple Storage Service (S3), API Gateway, AWS Data Exchange, and serverless models using Lambda functions. Marketing features of the Marketplace were captured, including tools for building awareness, promotions, and campaign tracking. An inventory of related and/or competitive healthcare offerings on AWS was also listed. The company proceeded to develop and release offerings in AWS based in part on this report.
