Enhanced Document Retrieval: A Government Sector Case Study

Overview

Time Under Tension recently partnered with a government client to create a cutting-edge web application designed to transform the way their staff interact with and leverage a vast repository of documents. This application is not your everyday document search tool – it leverages advanced AI to interpret and respond to user queries with remarkable precision and speed.

The Challenge

The task was multifaceted and complex. The client needed a system that could sift through thousands of documents efficiently, interpreting user queries and delivering succinct, accurate answers. However, it wasn't just about text extraction. The system had to maintain the integrity of the documents’ structure—headings, sections, lists, and tables—since this would be crucial in refining the search results and maintaining context in the answers provided by the Language Learning Models (LLMs).

Furthermore, due to the sensitive nature of the documents, data security was a non-negotiable priority. Ensuring that all data remained within the secure confines of Microsoft’s Australian data centres was paramount.

Our Solution

In response to these needs, we developed a dual-function web application. The 'search' functionality empowers users to pose questions regarding processes, procedures, forms, and policies. The LLM then processes these queries to deliver concise, precise answers, coupled with references and citations pinpointing where in the document the answers are drawn from.

Moreover, the application boasts a 'process' capability, an exemplary feature of which is the AI’s ability to identify and reconcile the usage of over 4000 acronyms within the organisation's documentation, flagging inconsistencies and differing definitions.

Throughout the project, the client engaged in rigorous testing, executing over 50 test cases on the LLM responses. Their satisfaction with the outcome was clear as they gave the green light for a second phase focused on enhancements and broader application rollout.

Technologies Used

We employed Azure Cognitive Search for its powerful indexing and search capabilities. For the AI component, we integrated Australian instances of Azure OpenAI, including versions GPT-3.5 and GPT-4, which resulted in significant performance gains due to localised deployment.

The application's real-time interactive elements are powered by NodeJS and Websockets, ensuring seamless user experiences. Data storage and management were handled via Azure Storage Blobs and MongoDB, ensuring swift access and reliable data integrity. Furthermore, Azure AD B2C was integral for secure authentication across the platform.

Summary

Our project with this government client is a testament to the power of AI in modernising document handling and retrieval processes. With a secure, AI-powered web application, we have delivered a solution that not only meets the client’s needs but also sets a new standard for document search and processing within the government sector. The success of the initial phase and the ongoing enhancements promise to usher in a new era of efficiency and accuracy in document management.