As J2’s Director of Application Development, I was excited to attend InterSystems Global Summit this year—especially with all the talk around AI. The reality lived up to the hype as Global Summit included many excellent demonstrations and seminars to this effect. One of the sessions I found the most interesting was titled “How We Used AI to Build Faster, Sell More, and Support Better.”
In this session, InterSystems Sr. Technical Online Course Developer Derek Robinson, Applications Developer Vivian Lee, and Principal Technical Specialist Chris Stewart showcased an impressive array of generative AI tools InterSystems has built internally to find new business. This team presented three examples: Video Storyboarding for Learning Services, A Bid Search tool for RFP responses, and a Customer Support Assistant named Yoky.
In the storyboarding tool, Derek Robinson demonstrated how sample resources loaded into private LLM can inform it how to respond to queries, including style guides, writing tips, samples, guidelines, and more. One of the key benefits is the reuse of artifacts already built to continually hone the efficiency of the AI system. Although this tool doesn’t fully automate the process of building storyboards for learning services (something also critical for AI applications in Healthcare), it was able to build much of the foundation of storyboards in the voices of the Learning Services team, enabling rapid progression through the beginning of the default workflow the team typically undergoes. One of the most interesting insights the team shared was the importance of restricting the output of the Large Language Model (LLM) to only one or two elements and allowing the team to build around that element.
Vivian Lee shared examples of using Vector Search in a bid tool used at InterSystems called RF-Eye. As you can imagine, InterSystems responds to many RFPs and it is much more efficient to have a computer do some of that work. In her first example, Vivian compared old searches using %SIMILARITY with new searches using Vector Search. This yielded many interesting examples, as the %SIMILARITY search simply returned data entries with matching words. The Vector Search, however, because it has use of an LLM, was able to discern the ‘context’ of the question and answer it appropriately instead of just matching on word elements.
To configure RF-Eye, InterSystems loaded many previous RFPs, which included question-and-answer formatted data. Because the Vector Database contained Vector converted data, this naturally mitigated data redundancy. Previously, data deduplication required identical questions. However, with the semantic understanding of an LLM, the system could naturally prevent semantically similar questions from being entered into the dataset for future queries. Amazing!
Finally, Chris Stewart introduced Yoky, the AI-driven Customer Support Assistant. Yoky has technologically impressive underpinnings, with a tech stack that uses Python in a Jupyter Notebook to ingest private documents, split them into chunks, and ultimately build indices to store in the Vector Store. Yoky is a classic RAG-Model enhanced AI Assistant—it takes a vector representation of a user’s question, uses it to semantically query the vector store created from the private documents, and sends the result, along with the question, and some instructions about tone to the LLM, which generates a context-specific response to return to the user.
This wasn’t the end, though, as Stewart showed this model following the REACT pattern to loop through the data up to three times to gather a response the system was confident in. All of the AI features here were facilitated by using LangChain, one of the most popular GenAI Python libraries available. This was a deeply technical example with truly impressive results.
At last, the team came together to show, at a high level, how to build these kinds of tools into your applications.
AI is here to stay and InterSystems is showing the path forward! We’re excited to look at new and innovative ways to include AI tools in the products we build either internally or for our customers, and to offer our customers a unique level of expertise in AI specifically as it is developed in the InterSystems space. Thank you to InterSystems and all of the presenters for making it a truly excellent Global Summit in 2024!
J2 Interactive is a strategic technology partner with over two decades of experience in healthcare. Reach out for a free, no-pressure strategy session, and let’s explore what the right tech can do for your enterprise.
