Artificial intelligence (AI) has become a hot topic, frequently making headlines for its groundbreaking capabilities — from generating lifelike images and videos to automating customer service at drive-throughs. However, beyond these headline-grabbing demonstrations, there are more practical and transformative applications of AI. One such use case is the creation of conversational assistants that serve as “knowledge navigators,” which function as sophisticated search engines capable of parsing an organization’s vast internal knowledge base. These tools provide users with summaries of relevant data and direct access to source documents.

Introducing ABC Assist: Navigating Knowledge Seamlessly

ABC Assist is an innovative AI tool designed to help employees tap into decades of trusted data across the organization. Built by the ABC’s Machine Learning (ML) and Artificial Intelligence (AI) team, with a user-friendly interface developed by the Innovation Lab’s front-end team, this tool has the potential to transform how staff access, explore, and utilize information.

The Why: Enhancing Strategic Workflows

ABC’s Corporate Strategy team, tasked with providing expert guidance across the organization, often needs to sift through hundreds of long, detailed documents. The manual process of extracting key information can be time-consuming, leaving less room for higher-value work such as analysis and decision-making. The need for a faster, more efficient way to research and aggregate information inspired the development of ABC Assist.

Building the Knowledge Navigator: How AI Meets RAG

At the heart of ABC Assist is a combination of a large language model (LLM), similar to the generative technology behind ChatGPT, and Retrieval-Augmented Generation (RAG). RAG enhances the functionality of traditional AI models by integrating authoritative, document-based data directly into responses. The system retrieves relevant documents from a vector database and seamlessly incorporates them into AI-generated answers, offering both a summary and a direct link to the original source.

This method reduces the “black box” nature of AI by providing transparency. Users can view the citations and underlying references that shaped the AI-generated response, ensuring accountability and traceability. It also allows for more tailored and accurate answers specific to ABC’s data, rather than generic AI responses.

Beyond Corporate Strategy: Broader Use Cases

While initially developed for the Corporate Strategy team, the Innovation Lab quickly saw broader applications for ABC Assist across the organization. For example, content creators could use the tool to navigate large archives of data, such as identifying episodes of a long-running program that cover specific topics. The system can dive into stored transcripts and deliver episode numbers and timecodes related to the queried topic, streamlining research processes for various departments.

Development Process: A Collaborative Effort

Development on the alpha version of ABC Assist began in January 2024. The Innovation Lab team decided to integrate the tool into the Presentation Layer (PL) codebase, which also powers the ABC’s public-facing websites. The PL is built on Next.js, using TypeScript, and offers a robust library of high-quality components such as text fields, buttons, and navigation elements.

To ensure a seamless connection between the backend and frontend, a WebSocket API was implemented, allowing real-time communication via standardized JSON messages. Users can interact with the system through a clean, intuitive interface, and provide feedback on AI responses via thumbs-up/down controls. Feedback mechanisms are crucial for continuous improvement and tuning of the system’s accuracy.

Once a user submits a query (e.g., “What are the implications of the Archives Act for ABC’s record-keeping practices?”), the system retrieves relevant documents, and a progress indicator keeps the user informed as the RAG process unfolds. Citations in the system responses allow users to click through to the original documents, promoting transparency and deeper engagement with the data.

Milestones and Lessons Learned

By April 2024, after extensive rounds of quality assurance (QA) and unit testing, the alpha version of ABC Assist was integrated into the Presentation Layer codebase and rolled out to a test group within the organization. Early feedback has been positive, and the tool continues to be refined based on user insights.

ABC Assist is a testament to the power of combining emerging AI technology with domain-specific knowledge to solve real-world business challenges. The collaborative nature of the project, with ongoing dialogue between the ML/AI team and the Innovation Lab’s front-end developers, underscores the importance of cross-functional teamwork. Regular communication and problem-solving ensured the project’s smooth progress, and both teams have played to their strengths to deliver a cohesive, effective solution.

The Road Ahead: Expanding Capabilities

While the current iteration of ABC Assist is a significant leap forward, the tool is continually evolving. Future updates include targeted document filtering to refine the retrieval process, and the ability to deliver streamed responses, improving the system’s responsiveness and accuracy. The ultimate goal is to roll out the tool more broadly across the organization, empowering not just strategists, but all ABC staff to access, analyze, and extract insights from the wealth of information at their fingertips.

By embracing AI and RAG, ABC Assist aims to reduce the time spent searching for information and increase the focus on high-value work, such as decision-making, content creation, and strategic planning. This project exemplifies how AI can go beyond buzzword status and deliver measurable benefits within organizations, setting the stage for more intelligent, responsive, and efficient ways of working in the future.

Conclusion: AI Driving Business Value

ABC Assist shows how AI can be used not only to automate tasks but to amplify human capabilities. As AI continues to evolve, its applications in the business world will become more refined and integrated into everyday processes. ABC Assist is just one example of how a well-crafted AI system can unlock the potential hidden within decades of organizational knowledge, providing real-world benefits for staff and enabling more informed, efficient workflows across the board.