Nudge

Revolutionize meetings with Nudge, the next-gen meeting app. Smarter, faster, and more impactful meetings start here.

Comment

Inspiration

Nudge was created to address a common challenge in meetings: inefficiency caused by prolonged discussions and unnecessary digressions. Weekly updates, club meetings, and scrum sessions often run over time, draining productivity, focus, and valuable resources. Wasted meeting hours translate to significant financial losses for companies and businesses—time that could be better delegated to critical tasks and strategic initiatives.

Nudge uses various measuring strategies including cosine similarity, vector embeddings, and context-window storing to assess on-topicness and relevance. By analyzing conversation flow, it keeps discussions focused, saving time, cutting costs, and boosting efficiency.

What it does

Nudge offers a convenient meeting solution designed to enhance efficiency by analyzing speech in real time to determine if discussions are on-topic, issuing warnings when necessary, and providing statistical insights to further improve meeting productivity.

Users can schedule meetings, define time blocks, and set specific topics for structured and efficient discussions, ensuring every meeting stays focused and productive.

How we built it

Built with peer-to-peer services and WebRTC, it streams audio batches to a Flask backend via WebSockets. A locally embedded Llama 3 instance is then used to generate sentences that are very likely to be included into the meeting. The information is then used by our sentence transformer to produce a vector which represents the sentences in a mathematical way. These vectors are then compared together using cosine similarity which determines the on-topicness of the users. Utilized firebase for meeting/session management, React TS frontend.

Challenges we ran into

We first attempted to construct the platform with mediasoup, but retrieval and sending of RTP packets from a node backend to flask backend was far too difficult (we did not want to touch udp). In addition, mediasoup proved difficult in learning completely in a short time frame.

Accomplishments that we're proud of

Our team is proud to have created an application that is useful and complete. It combines multiple machine learning models capable of being hosted locally in order to deliver a productivity boost that users strive for!

Built With

Share this project:

Updates