Beacon Pitch - UCL First Response in a Box Design Hackathon 2023
0Beacon Pitch - UCL First Response in a Box Design Hackathon 2023
Online and offline AI models
Why is it useful?
According to the National Geographic, on average 10,000 people die in earthquakes annually. The majority of deaths are caused by collapsing buildings, but mud slides, fires, floods, or tsunamis often exacerbate such situations. Our idea, Beacon, focuses on prevention, by decreasing the scale of damage caused to human lives, but also assisting victims in the aftermath of such events.
Beacon consists of a user-friendly application that assists the user in need and also a earthquake detection system to be used by emergency response teams. The application assists citizens through the provision of a chatbot which can answer multiple queries. For example, what the nearest safe spots (similar to hotspots) are in their area or how to administer emergency first aid. It also has an emergency SOS feature and a ‘Find your peers’ service. This service will display where people have congregated in the event of earthquake helping the user navigate to a safe place.
Our chatbot’s primary target market is people living in high catastrophe risk countries, especially where access to a stable internet connection is limited and other infrastructure is limited, such as in LICs. The chatbot uses a GPT-like offline model that is local to the user's phone.
The emergency detection system aims to help with preventing the scale of damage. Many phones and even laptops come equipped with accelerometers, which have the ability to detect unusual vibrations. These sensors are sensitive enough to act like mini seismometers. Our system will combine data from multiple devices, such as phones and laptops, to calculate the location of an earthquake event. If the vibrations are being detected by 20+ users within a certain range, then the app will automatically identify that as an earthquake. This will help decrease casualties as even a few extra seconds can allow citizens to undertake preventative action, since radio signals will transmit faster than the seismic activity causing the earthquake. For example, turning off public transit to avoid derailment, or preventing damage to power, gas and other networks than can lead to secondary risks such as fires. Data collected from these devices can then be used to analyse trends and patterns to predict future earthquake events.
Unlike existing earthquake detection systems, Beacon’s USP is that it takes advantage of forming peer-to-peer networks using devices we all carry around in our pockets. This expands our target market to areas where complex and expensive detection systems are unavailable, such as poorer regions in the world.
Our app connects to nearby phones via WFD to ensure offline communication. When the connection is established between phones, they share data (from the earthquake detection and GPS location) with each other. When a single phone from the network obtains an internet or satellite connection, it instantly sends the data to the rescue team or organization. They can then use this information to help those in need.
Summary:
Used online AI model: Online OpenAI’s GPT-3.5
Used offline AI model: Local version of OpenAI’s GPT-2
Model configuration and usage:
LangChain will be used to train both our GPT models on data such as reports from specialists in disaster relief and solutions used to encounter such situations. We will use an appropriate sample size and test for consistency.
What data is needed?
From the user we will be collecting essential medical information such as their blood type in order to advise them on conducting first aid in emergency situations and for the emergency response teams,
Data collected from user’s phones and laptop will be used to detect seismic activity. Changes in the vibration of P-wave characteristics will help determine earthquake events.
We will also require demographic data of different locations, road maps and safety routes. In order to train the chatbot on first aid recommendations we will use approved medical advice.
References:
Vels, E. (2023). Tech solutions help limit the aftermath of an earthquake. [online] Innovation Origins. Available at: https://innovationorigins.com/en/tech-solutions-help-limit-the-aftermath-of-an-earthquake/.
Ough, T. (n.d.). The phones that detect earthquakes. [online] www.bbc.com. Available at: https://www.bbc.com/future/article/20230405-the-phones-that-detect-earthquakes.
NPR. (2010). Your Laptop Could Detect The Next Earthquake. [online] Available at: https://www.npr.org/2010/04/17/126073353/your-laptop-could-detect-the-next-earthquake#:~:text=Laptop%20accelerometers%20aren [Accessed 1 Oct. 2023].