Country Fire Service of South Australia (CFS) is a highly visible government agency dedicated to protecting the lives of all South Australians and their property from the ever present dangers of fire.
With a commitment to providing fire and emergency services to South Australians, the CFS aims to educate the public on fire prevention and safety management.
In a step to further enhance its communication and educational efforts, the CFS embarked upon an AI pilot using the New Dialogue AI platform. New Dialogue AI is based on an open-source Large Language Model (LLM) which enabled the CFS to implement the platform within their existing cloud tenancy. The end result was the implementation of a public facing AI chatbot presented within the CFS website.
The objective of the pilot was to;
New Dialogue AI was chosen for its ability to run securely and privately within SAFECOM (South Australian Fire & Emergency Services Commission) infrastructure, and its ability to process and generate human-like text responses, making it a suitable tool for creating a contextually relevant and effective chatbot.
The AI chatbot was designed to provide real-time answers to user queries, and enabled the audiences to access accurate and relevant information concerning fire prevention and safety management practices quickly and easily.
To achieve this objective, the New Dialogue AI was provided access to a dataset of fire prevention and safety management knowledge, allowing it to generate accurate and relevant responses to user queries.
The New Dialogue AI chatbot was designed to be user-friendly, through providing an intuitive user interface that made it simple for external users to pose questions and receive timely responses.
The pilot was designed to provide user-friendly access for audiences to chat with the CFS website. Clicking on a dedicated chat icon within the bottom of the CFS website launched the AI chatbot. Once clicked, users were presented with a familiar chat window where they could enter text queries using natural language.
The New Dialogue AI then processed these queries in real-time, with the aim of generating contextually relevant responses that were displayed back to the user via the chat window.
The LLM was deployed within existing infrastructure, and this allowed for seamless integration with the CFS website which ensured the chatbot could handle high volumes of traffic without any performance or degradation issues. The LLM was also configured to allow for real-time monitoring of the chatbot’s performance when multiple sessions were active.
Deployment of the New Dialogue AI leveraged existing AWS infrastructure managed by SAFECOM. The specifications for the LLM environment were;
Amazon EC2 G5 instances are the latest generation of NVIDIA GPU-based instances which can be used for a wide range of machine learning use cases.
To test New Dialogue AI against what could be a wide range of natural language queries, a suitable dataset of fire prevention and safety management content was shared with the LLM and through RAG (Retrieval-Augmented Generation), a process to optimise an LLM so it references an authoritative knowledge source outside of its baseline training data, contextually relevant responses were possible.
This knowledge base included questions related to fire safety regulations, fire prevention techniques, emergency procedures, and more. The LLM was able to generate accurate responses to these queries by analyzing the input text and selecting relevant answers from its training data.
During the pilot the AI chatbot received over 600 queries from the audiences. In general, queries posed by the audiences covered a wide range of topics on fire prevention and safety management, how to action something and also included queries on fire regulation, emergency procedures and fire prevention techniques. A gist of the most common queries were centered around;
And the longest query;
"I need to ring region 5 and ask them if they can send someone out to my property as soon as possible because we have a fire that is getting out of control and we need help immediately and i don't know what to do and i can't find any information about how to contact them directly and i'm really worried and scared and i need help as soon as possible"
New Dialogue AI was able to provide contextually relevant responses to queries prompted by the external audience, and provide these in real-time. The pilot demonstrated how a LLM can be used at the frontline of contextually relevant information retrieval and delivery.
The AI chatbot’s user-friendly interface made it simple to implement within the existing CFS website, but also very accessible for the external audience to discover it, ask questions and receive answers, while real-time monitoring ensured any performance degradation was quickly identified and addressed.
Overall the aim of the pilot was to demonstrate the potential of New Dialogue AI, and how open-source Large Language Models can be used safely in context to accessing a controlled dataset of fire prevention and safety management knowledge for the benefit of improving response times to queries posed by the public.
By providing users with quick and easy access to contextually accurate information, the New Dialogue AI chatbot demonstrated a feasible approach for accessing and promoting fire safety awareness for all South Australians.
