Automateam

All meetings participan(present in the transcript) according to duration spoken
Topics detected and how the relevancy votes look like.Topics are simulated for the backend (didnt have time to integrate with python)
Add new meeting transcript according to Msft team transcript specification.
Main view, all meetings related to a project. Clicking view topics shows all topics detected for the meeting.
All meeting participants(present in the transcript)
Summary of the meeting

Inspiration

Solving everyday pain that millions of meeting participants suffer when they have to take notes, while trying to make impactful decisions.

What it does

It facilitates notetaking by summarizing Microsoft Teams transcript and extracting topics discussed during the conversation. Automatic email sending with meeting notes and topics discussed to meeting participants. It keeps track of all meetings related to each project in a timeline-like manner. Allows filtering by topics discussed to allow simple tracking of meeting followups. Enables further improvement by allowing participant to improve our summarization model while it is being used. Shows minutes that each meeting participant spoke.

How we built it

Our solution is split into 3 components. The front end is built on React and deployed to Heroku. The backend is built on Spring Boot and H2 database and deployed as an Azzure App Service. Our python endpoint which uses Transformers and Bert methods for summarization and keyword extraction respectively.

Challenges we ran into

Biggest challenge we faced was the inability to deploy our python service in the cloud due to very big size of libraries used. We didn't have time to optimize the dependencies and as a result we had to "prototype" our integration between backend and machine learning service. Fortunately they both work independently :)

Accomplishments that we're proud of

Very good team work. Very clear separation of responsibilities within the team. Each one of us made the most use of our abilities.

What we learned

Learned a lot about Azurre and especially about Team Apps although we didn't see it reasonable to use some Team Plugin/Extension for our purposes. Most challenging part was definitely summarization problem and topic extraction, especially with the few amount of data we have.

What's next for Automated Microsoft Teams Notetaking

We would like to further improve our model and make it able to learn. We already provided the feature that meeting participants can vote/ help train the model by agreeing or disagreeing with its conclusions, however we have not yet found a way how to feed this useful information in our model. Gathering more english meetings transcript. Of course finally deploying our Python API and allow full integration between our components.