AI-Powered Software Company Uses Generative AI to Update Training Materials
Executive Summary
The AI-powered software company partnered with Mission Cloud to automate the creation of training course templates for healthcare professionals using generative AI. The solution, developed with Amazon Bedrock and Anthropic Claude 2, streamlined the processing of videos, PDF files, DOCX files, and images, thus speeding up the development and delivery of structured training programs.
Amazon Titan Multimodal Embeddings G1 enabled the conversion of images into searchable vectors, which allowed for the extraction and retrieval of relevant images from source materials. The customer now has access to a generative AI solution that streamlines the training material development process.
About the Customer
Background: Frequent Updates to Critical Information
The AI-powered software company allows its customers to rapidly create training materials that improve clinician proficiency when using the Epic EHR system. Its clients have to make frequent updates due to regulatory changes, internal workflow customizations, and quarterly software releases. Detailed guidelines for EHR processes in the healthcare industry are crucial, and the speed at which updates are released is critical.
The software company wanted to use generative AI to streamline the development of training courses and programs. As an existing user of Amazon Web Services (AWS), the customer wanted to keep its hosting strategy within the AWS ecosystem. After meeting the Mission Cloud team at a community-building event in Washington, D.C., the customer asked for help developing a solution that showcased the potential of generative AI in automating training course template creation.
The Challenge: Accelerating the Training Material Development Process
The customer wanted to use generative AI to quickly create and deliver more structured training based on subject-matter documents related to specific learning topics. These source documents included text descriptions of Epic EHR workflow procedures for specific roles, as well as videos.
The goal was to identify the extent to which generative AI could streamline the development of training courses and programs. The customer aimed to automatically generate elements within a training course template, including course titles, descriptions, and learning objectives. Additional elements included course outlines, text-based lesson content extracted from source documents, and relevant assessment questions.
Solution: Using Generative AI to Streamline Production
With a customer-driven focus on prompt engineering, Mission Cloud developed a command-line interface tool using Amazon Bedrock and Anthropic Claude 2. While the tool had a simple interface, it incorporated various pipelines to process data and generate results. Amazon Titan Multimodal Embeddings G1 enabled the conversion of images into searchable vectors, which allowed for the extraction and retrieval of relevant images from source materials such as PDFs and DOCX files.
The solution addressed four primary goals: processing videos, processing PDF files, processing DOCX files, and processing images. Initially, each goal was approached independently, with standalone Python modules developed for each effort. The focus was on ingesting files in different formats and transforming them into the desired output format. As the project progressed, these individual modules were combined into a cohesive pipeline. This pipeline created a single flow for Mission Cloud to iterate over files contained within a directory, combining them to generate the final course content.
The solution contained two pathways, with the first translating a single document, such as a facilitator guide, into a single course. The second pathway bundled multiple documents of different types, such as PDFs, DOCX files, and videos, and transformed them into a single lesson plan. Training material could be fed into Bedrock, where documents were analyzed and transformed into a training course template in XML format.
Results: Access to Tailored, Up-to-Date Training
Mission Cloud successfully achieved the customer's initial goals ahead of schedule and within the planned budget. Once implemented, the solution should complement the customer’s revenue-driving capabilities by simplifying training course updates. The customer gains a competitive edge and demonstrates its commitment to enabling clients to provide valuable, relevant, and up-to-date training to their end-users.
“Instead of simply providing prompts in a Word document format, Mission Cloud took the initiative to thoroughly understand our use case, ensuring the desired result was achieved. [Mission Cloud] engaged in numerous touchpoints with us, seeking feedback and iterating multiple times to ensure a successful outcome.”
– Chief Information Security Officer at the customer organization.
The healthcare education company continues to work with Mission Cloud for ongoing experiments and considerations, such as the exploration of opportunities in transcription and search capabilities.
AWS Services Used as part of the solution
During the engagement with Mission Cloud, the customer used a variety of AWS services, including:
| Amazon Bedrock |
| Amazon Titan Multimodal Embeddings G1 |
| Amazon Transcribes |
| Anthropic Claude 2.1 |