I’m Maggie Corry, chief engineer at Booz Allen. In partnership with industry leaders, we deliver ready-to-deploy customizable Generative AI solutions tailored to client needs. Today, so much data is scattered across different formats with different jargon and terminologies. Our GenAI for data processing solution allows teams to make use of their messy, disparate data by automating how they extract, connect, and synthesize it. This improves accuracy, enhances interoperability, and streamlines workflows. Users can upload their data and ask plain language questions. Connections and insights that may have taken days are now found by agents in minutes. Let’s dive into how the solution works using a health data use case.
First, our GenAI solution ingests data from a variety of sources and formats, including structured, unstructured, and multimodal data. In this scenario, we have unstructured text in PDF health data, including doctor’s notes and lab test results across multiple patients. The solution creates a knowledge graph showing connections across the data that we’ve uploaded. To do so, it converts that data into a format that AI models can understand while preserving the original meaning, context, and nuance. Next, it identifies key entities such as patients and symptoms, their attributes, and relationships across the data. It also standardizes this data, such as identifying that hypertension and high blood pressure reference the same thing. The system links these entities to trusted knowledge sources, uncovering connections and contacts that might go otherwise unnoticed. Now you have structured and connected data that meets defined specifications and is ready for further analysis. Looking closer at this knowledge graph, you can see the complex relationships it has extracted. Let’s focus in on one cluster of the knowledge graph, a gene.
This gene has been linked to different entities such as lab test results, patients, and several different conditions such as Type 1 Diabetes. While this may take doctors days to piece together, our solution takes minutes to identify anomalies and patterns. The summaries tab has summaries of these findings. You can analyze this data on identified clusters, such as for patients with diabetes-like symptoms. The search tab allows you to insert queries such as SPARQL to better understand the relationships across the data. You can also use the chat feature with agenetic GraphRAG capabilities to carefully traverse the data mapped out in the Knowledge Graph. Here I’ve asked a question about the gene and patients we just looked at. As you can see, the agents have tied together the disparate notes in lab data to give a comprehensive analysis of the patients, ultimately reducing exploration time for a doctor.
You can see in detail how the agents got to their answers to give confidence that the model is not hallucinating, which is a risk with generic chat bot implementation. Our GenAI for data processing solution makes sense of messy, disparate data. It automates the extraction, connection, and synthesis of information, turning scattered data into structured, useful insights that improve accuracy and streamline workflows. This saves time for any analyst providing them with insights faster. This solution works with various data types like chat logs, JIRA tickets, emails, meeting transcripts, and much more. It’s customizable and integrates seamlessly with your existing cloud infrastructure through one-click deployment. Booz Allen’s GenAI for data processing—accelerating outcomes with unprecedented speed and efficiency.
