hi I'm luk anapoli industry Solutions consultant on Long following out on our previous video discussing image search for claim processing we'll see how insurance companies can Implement retrial augmented generation or simply rag to accelerate and improve claim adjustment operations rack is the process of providing contextual information on to large language models or llms and enabling them to give us better answers that incorporate proprietary data or knowledge that were not included in the raining Corpus the diagram illustrates a high level workflow and the essential Architectural Components required to build a rug application initially the user prompt under goes vectorization followed by embedding into mongor Atlas Vector search retrieves similar vectors which are then used to retrieve the most relevant documents from our collection once we obtain the documents we fit them to the lln together with the original question consequently the AI generated answer is enriched by both the user prompt and retrieve documents resulting in a significantly more informative and Tor response compared to generic model now let's dissect the process into its individual steps and see what actually happens in practice the context as in the previous video revols around claim adjustment or specifically car accidents here the adjuster's task is to probably assess the damage incurred and determine the appropriate monetary compensation for the customer this assessment involves comparing the current claim with past ones in our example the user is particularly interested in claims resulting from adverse weather conditions The Prompt is embedded and pushed to atas as mentioned earlier the relevant documents are retrieved and now we g a clear understanding of what relevant entails in this context highlighted in Green in the user's question we see a reference to adverse weather in the document Reed by Vector search we notice a mention of heavy rain in the caim description field this is the key point of rag preserving the semantic meaning of the query we're not looking for an exact match as we would expect from a regular database search Vector search returns vectors that embeded conceptual similarity this marks a significant paradig shift now the relevant documents are fed to the LM as evident in the response the adjuster is provided with a summary regarding weather related accidents encompassing dimensions of rain hail and fire essentially the user is talking in natural language with a data pretty cool huh in a moment we'll see how this over a in a business application but first we have some bonus content to share with you we built a nice UI for image search now user can simply drag and drop the query image to click Rel similar photos and Associated claims in seconds all to the right down on now on the left we have a chat box where users can type questions or choose from predefined ones let's ask about weather related claims and their average loss amount highlighted in Gray we see the answer provided by by the llm that includes a calculation of the average loss amount for weather related claims on the right side we can see all the documents used as context by the AI as we can see the claim description field contains references to hail storms and thunderstorms in addition to the description other fields are also included providing the claim adjuster with a comprehensive overview of relevant claims this facilitates a quicker estimation of the damage for the active claim the implementation of rack represents a significant advancement for insurance companies enabling them to access damage more rapidly and accurately this in turn translates into an enhanced customer experience in summary the fusion of mongodb and llms is a game changer for claim processing as we just seen finding relevant information is fast and convenient LMS are able to answer a wide range of questions reducing the need for upfront planning and system design with their ability to understand natural language user interaction is intuitive and doesn't require any special system skills but above all we're finally able to access and leverage as fracture data all those PDFs photos or videos that were buried somewhere in a data Lake can finally come to life and help us streamline business processes claim handlers Underwriters or customer service operators can all benefit from rack powered applications and services most of the time the same data model can serve multiple personas if the right data strategy is in place feel free to reach out to the industry Solutions team to learn more about that thank you for watching