Make Pets Found (using AI) - Day 1
Preparing for takeoff
The Gravitywell team gathered at Bristol airport, eagerly anticipating a smooth, late afternoon flight to Malaga. To avoid any mishaps, Laura had suggested allowing at least 72 hours to get through security and find our gate. You can never be too sure. Alas, the flight was delayed! Two-and-a-half hours (and a handful of Double Whoppers™) later and we were finally taking off, leaving the rainy West Country skies in our wake and gliding towards the south coast of Spain.
Our journey seemed somewhat cursed however, as upon arrival we unwillingly embarked on a car rental egg hunt. Even Jesús, our Andalusian native, was confused by the lack of signage. Slightly worrying. Eventually, after a tour of the various levels of Malaga airport’s car park, we located our rental company. Off we went, Simon and Luke the designated drivers speeding off into the warm night towards a carefully selected mercado, to pick up some essentials (beer, wine, etc.) for our first night in the hills.
Supermarket sweep complete we headed north, arriving at El Molino Del Conde around 1am local time. We were greeted by a cute little farm cat who helpfully showed us to our rooms, then not-so-helpfully relieved himself in the brick oven. Simon eagerly offered to clean up the mess. After a nightcap of chorizo and cheap lager we all turned in for the night.
After waking to the sounds of the Andalusian wild, we met for a stand up (well, a sit down) in the kitchen. We quickly realised that a huge challenge for us this week would be internet connection. The speed is 1/10,000th of the speed that we’re used to in the studio. This is something we’ve encountered on previous hackathons however, so we’re used to having to adapt to less-than-ideal situations!
While Simon, Laura and Hugo went to the local supermarket for supplies, the rest of the team cracked on with their various tasks. Our goal by the end of the week is to build an app using machine learning models that will be able to compare images of pets and match them together correctly.
Jesús had done a load of research and prep prior to the week. He presented his findings to the rest of the devs, who then did more reading into high level machine learning concepts, such as use-cases for supervised / unsupervised training and deep learning. It’s a huge field so we had to leave out some of the more complex mathematical detail, but it gave us a good enough understanding for the rest of the week.
Theory done, we then got started by looking at implementing Google Cloud Vision to detect what data is in the uploaded image and retrieve basic labels. Next, we had to understand and normalise the data that we got back from the API into a format that we’d be able to use in our app.
Meanwhile, Sam and Henry were getting set up with Jupyter locally so that we could experiment with some machine learning models and begin training them. James began setting up an Expo app for a cross-platform frontend, whilst George scaffolded a backend API using Mongo for a data source.
Whilst the aim of the week is focused on research and experimentation, Matt and Simon began to think about the future vision for a user-friendly product, putting together some high-level wireframes.
We rounded off the evening with a delicious meal courtesy of Henry — ‘Turkish beans with a Spanish twist’ (his words), and a few too many cans of Cruzcampo.
Dia uno: completado.