MeetingMind by Cactus - AI Solutions

03. SOLUTION

Leverage deep learning

The technical solution for MeetingMind consists of leveraging the capabilities of three different neural networks:

faster-whipser for the transcription task
pyannote.audio for speaker diarization
META llama3 70B for text based tasks

Due to data-privacy reasons, the first two neural networks are self-hosted. For testing purposes we opted for the META llama3 70B model hosted by www.groq.com. This model could be substituted by a LLM of choice, either self-hosted or by using cloud providers such as openAI or Anthropic.

The only required input is an audio file which contains the audio of the meeting recording.

04. RESULT

Fully integrated in the meeting environment

The result is an app integrated in Microsoft Teams (the main communication tool in our company). It can be added to an online meeting. The meeting is recorded using the Teams built-in recording function. Once the recording is available for download, the app notifies the meeting attendees and the audio transcription is stared upon user input. The output format of the meeting notes can either be chosen via a set of predefined template depending on the meeting setting, for example “weekly meeting”, or individually specified by user input.

Make meetings more efficient and traceable

01. Intro

02. Challenge

03. Solution

04. Results

01. INTRO

Make meetings more efficient and traceable

02. CHALLENGE

Translating Social Impact into Actionable Insights

03. SOLUTION

Leverage deep learning

04. RESULT

Fully integrated in the meeting environment

CONCLUSION

Increasing efficiency and traceability of online meetings

Televic

Cardoen

Xandres