As a Microsoft Gold Partner, Screenmedia had the pleasure of being invited to the tech giant’s AI hackathon last week, focusing on pushing the limits of our natural language and search capability.
Joining nearly 20 other Microsoft partners at the Microsoft Reactor in Shoreditch, the day kicked off with a showcase of the current state of AI at Microsoft, before sinking our teeth into the capabilities of the Language Understanding (LUIS) cloud-based API service.
Our aim during the hackathon was to explore the capabilities of LUIS in a travel context; as it turns out, asking how to get from A to B can cause more problems than expected, especially when you throw obscure Scottish towns into the mix. We attempted to extract timetable information from PDF documents into a useable format, but while the extraction process itself worked well, the resulting format of the data still made it hard to interpret as a timetable and we ended up manually processing a few of these file to help us move on.
[Credit: Adam Jackson, Twitter]
Our problems with geography did not get any easier as the hack went on; while we could reliably work out where we were going from and to, our source data didn’t have ‘Glasgow’ or ‘Aberdeen’ as terminal locations for any travel route. The next challenge was how to take general terms for a location to find specific points of interest and calculate routes. Azure Maps came to our rescue here; with some well-defined geometries and search features we were able to arrive at a solution to this problem.
Last on our list was to focus on time and route information. LUIS quickly helped us with time ranges for ‘tomorrow’, ‘next week’, ‘next Sunday’ and asking for things like ‘first’ or ‘last’. We combined this with our timetable data and were quickly answering questions such as “When is the last bus from Glasgow to Aberdeen on Sunday”. We started to explore the wider routing and connections problem with graph databases but unfortunately there’s a limit to how much can be achieved in three days!
The event also featured lightning talks from both Microsoft specialists and other participating teams, covering updates and bests practices around machine learning tools, the operationalisation of AI, as well as some of the latest features in Power BI. One team presented promising work using Azure Machine Learning to control rudder control systems for racing yachts, while others looked at text extraction and encoding for medical notes, safeguarding protection of school-owned communications to identify potential abuse, harassment or bullying, and analysing contract documents to identify compliance with the placement and number of signatures and dates.
The hackathon provided valuable time to focus on expanding our knowledge in a cutting edge area with direct support from experts at Microsoft. We’re looking forward to being invited back in future, and hope to bring along one of our clients, as there’s nothing we love more than innovating to solve real world problems!
We run our own hackathons and design sprints to tackle challenges our clients are facing. If you have a tech, strategy or business problem you're facing, get in touch and see how we can support you.