La Girafe

our team hard at work preparing the final presentation

Inspiration

We wanted to leverage machine learning and other cutting-edge technologies to help our client work more efficiently and leave early to go watch the YB soccer match.

What it does

It ingests tickets (that come from mails, but could be other sources) and categorizes them roughly so that we know who's best able to answer them. Furthermore, it provides detailed suggestions based on the content.

How we built it

We first talked to the stakeholder and defined our target audience. We then methodically thought about the pain points that we wanted to solve and came up with a user experience that would help them. After that, we quickly set up a first deployment pipeline to always have a running version of the product. For this we used GitHub and Azure DevOps pipelines. We deployed the artefacts to Azure App Services, the main one being a Flask web app with a Vue front-end. During the next 18 hours we added additional APIs to that app that fetched real data using different machine learning services that we aggregated with Flask. During all this time we were able to debug locally, which is a big advantage compared to, say, Azure Functions.

Our experts on natural language processing evaluated different Azure Cognitive Services APIs, then our web app implementors called the APIs in a secure way and exposed them in a nice UI.

Challenges we ran into

We obtained a data set from our client that we used to teach our cognitive services. However, we will need a lot more data to train the model for real use.

Accomplishments that we are proud of

We managed to translate the requirements of the client into a working solution in a very short time. We were able to methodically analyse what was needed and use the different expertise of our colleagues to have a product in the end that illustrates our solution with real code.

What we learned

We were able to write our first cloud native application! Also, we were able to use methodologies like Design Thinking to rapidly converge on a UI design. We learned to appreciate the breadth of possibilities in automated machine learning in Azure (out-of-the-box solutions as well as custom adaptations) to help in practical tasks and bring actual value to our client.