Scouting: Simplify the Reading and Aggregation of Information with Gen AI

The challenge

A common problem for many companies that manage large amounts of data is the ability to access information easily and directly. This obstacle can arise from a number of factors, including the integration of heterogeneous systems, the lack of adequate analytical tools, and difficulties in ensuring the quality and reliability of data.

Companies often use multiple software programs to manage data. This makes the process inflexible, requires knowledge of different tools, and limits the accessibility of the data itself.

Needs

Scouting's needs were to have an information aggregator that would help the operatives to extract and re-elaborate the data in a simple way. A tool that would allow to obtain timely answers to specific questions, supporting the onerous point-by-point analysis of the large amount of data available.

QueryBot: Using Natural Language to Analyse Business Information

Claranet's proposal to meet Scouting's needs was to implement an AI-powered QueryBot. This powerful tool offers a new way of interfacing with business services through natural language.

The chatbot, before responding, will search for the necessary information by contacting the services and will process it to provide a response based on the text input provided by the user.

This technology offers additional significant advantages. First of all, there is no need to learn new skills to use it, which makes it accessible to everyone. The learning curve is low, allowing any type of user to adopt it easily. Furthermore, thanks to the ability to specify the output of the results for a specific need, manual operations are minimised, making the process more efficient and targeted.

Generative AI and Enabling Technologies

Generative AI, Prompt Engineering and RAG

But how does QueryBot work and what are the enabling technologies?

QueryBot uses Generative Artificial Intelligence (Generative AI) to understand user prompts and answer their questions, leveraging techniques and models that can generate textual responses.

Although generative AI aims to mimic humans, it requires precise instructions to produce relevant and high-quality results. Prompt engineering is the process of guiding Generative AI solutions to produce the desired results. The most appropriate formats, phrases, words, and symbols are identified that guide the AI ​​to interact with users in the most meaningful way.

At this point it was necessary to provide the QueryBot with a solid knowledge base related to the domain and the Scouting knowledge base. Through Retrieval-Augmented Generation (RAG) it was possible to extend the capabilities of the LLM to enable the QueryBot with organisational information and response optimisation.

Infrastructure

To ensure efficiency and scalability, we managed CI/CD through Github Actions, which automates the build and deployment process on AWS. Containers are stored in Amazon Elastic Container Registry (ECR), while querybot instances run on Amazon Elastic Container Service (ECS). Additionally, we integrated Bedrock to use large language models (LLMs), improving the system’s understanding and response capabilities.

Result

With our support, Scouting is now testing a new way to access and analyze the vast amount of data available. Our QueryBot is trying to approach a new way to simplify the operations of reading and aggregating information, allowing Scouting operators to "chat" naturally to get what they need. We are collecting feedback and evaluating the performance to ensure we provide an effective and reliable solution.

Find out more about Claranet's Data and AI solutions, and speak to one of our experts today.