Meetsy is a meeting assistant that transcribes, summarizes, extracts actions items from meetings and also makes it all searchable so you don't have to go through the whole meeting again to find that one thing you were looking for.
This repository contains the code for the AI backend of the project. The frontend can be found here and the demo can be found here.
The app uses the CellStrat Whisper API to transcribe the meetings. You can sign up for a free account here. Once you have an account, you can get your API key from the deployment dashboard.
The Insights API is responsible for the Meeting Summarization and Action Item Extraction. Both of these tasks use the GPT-3 API from OpenAI. This part is setup as a lambda function on AWS for additional post processing. You can find the code for the lambda function here and the lambda layer setup for the openai package here.
- Follow the instructions in the lambda_layer's readme to setup the lambda layer.
- Create a new lambda function on AWS with the following settings:
- Runtime: Python 3.8
- Handler:
lambda_function.lambda_handler
- Layers: The lambda layer you created in step 1.
- Environment Variables
- OPENAI_API_KEY: Your OpenAI API key
- Copy the code in insights_api folder to the lambda function.
- Create a Lambda Function URL to invoke the lambda function from the frontend. Make sure to apply the appropriate CORS settings and authentication to None.
- Don't forget to increase the timeout of the lambda function to 2 minutes or something similar.
The search API works on the transcript of the meeting to find the relevant timestamps for the search query. It uses Sentence Transformers to find the most similar sentences to the search query. This model is deployed on CellStrat Hub as well who's code can be found at transcript_search.
- Open your workspace on CellStrat Hub and upload the transcript_search folder.
- Open up a terminal and run the following commands as a pre-requisite to install the required packages:
cd transcript_search pip install sentence-transformers python download_model.py
- Now build and deploy the model using the following commands:
hub build hub deploy
Learn more about deploying models on CellStrat Hub here.