AWS reInvent. Some sessions
Alexa in Higher Ed
Amazon has run some contests for students: Hack-the-Dorm and Hack-the-World
John Rome, Deputy CIO of Arizona State presented. ASU has over 100,000 students. Aiming to become first voice enabled campus.
ASU gave 1,500 Engineering students in one residence Echo Dots and built a skill for the campus: events (gigs, sports), business hours for facilities, academic calendar (start of term, end of term etc). Really popular with students: 80% satisfaction, 63% use Alex every few days or more (not just for the ASU skills, timers and music are popular).
Data shows that events are the skill most used by students. Fortunately ASU already had an events API they could build on to make this.
Then they asked students what else they wanted next:
Other ideas included directory search and parking, chatbots that link with Slack for helpdesk, prototype integration with digital signage.
To deliver these next services, hey need to build APIs for some and work with third parties for others. Just to add to complexity, it’s not just the usual enterprise vendors that they need to engage e.g. catering franchises need to expose their data to populate menus.
The tech stack is pretty standard: Alex voice service + other Alexa backends, API gateway, Lambda, S3 and a VPC back to on-campus systems.
The second part of the session was Canvas talking about their Alexa skill. They presented stats that showed adoption is broad but not deep yet (a few thousand invocations from over 400 sites). Star ratings for the skill also mentioned: heavily skewed between 1 star (couldn’t get it to work) v 5 star (people who could get it to work really liked it). Next steps are to work with institutions to push the skill and drive usage.
And finally, Canvas have found that the most enthusiastic region in the world for Alexa is the UK!
Machine learning in commodity markets
Commodities are physical, and lots of paper documents used to track ownership and location of assets. This creates risk as manual progress needed to reconcile position.
IHS Markit have developed solution to digitise documents, scape data, audit and disseminate the data. They take the documents then use ML to classify the document into type e.g. bill of loading
Further stages then do OCR for the different types of form to create inventory tracking.
They use the Amazon ML AMIs plus AWS compute with GPU, and have a bot that uses Amazon Lex for the natural language processing.
There’s not a lot of detail in this writeup as there wasn’t a lot of detail in the talk!
