tweag / work-daigest

Create a digest of your work week using a LLM

Summarize your work using a LLM

This project came into existence as part of a Tweag-internal hackathon on GenAI topics. The goal is to summarize your work using data from multiple work-related sources, such as

• GitHub,
• Google Calendar,
• Slack,
• Jira,
• ...

Right now, supported data sources are GitHub and Google Calendar. With these sources, the program yields a nice summary text, for example (in the UI):

Usage

Configure AWS Bedrock and AWS credentials

The application currently calls out to AWS Bedrock for LLM access. So you'll have to enable the relevant models (default is Claude-3 Sonnet) in Bedrock.

Once that is done, make sure that you have local AWS credentials with all necessary permissions set up, for example using aws sso configure and aws sso login. Don't forget to set the AWS_PROFILE environment variable to your AWS profile name if it's not the default.

Configure GitHub (optional)

If you want data from private GitHub repositories be included in the summary, you need to set up a GitHub personal token. The program expects a (classic) GitHub personal token in the environment variable GITHUB_TOKEN. That token needs to have the full repo OAuth scopes.

Set up the software environment

To get started, install the program in a virtual environment using nix-shell if you're a Nix person. If you're not, you'll have to have Poetry installed. Then, you can just run poetry install to install the application and all required dependencies and you're ready to go.

Get the data

GitHub data

GitHub data (issues, PRs, commits) is retrieved automatically using the code in work_daigest/fetchers/github.py, but follow instructions in the fetchers README if you'd like to manually fetch the data and inspect it.

Google Calendar data

Currently, Google Calendar data needs to be exported manually into an .ics file. To do that, open Google Calendar in your browser, then go to "Settings" (the cogwheel symbol) -> "Import & export" and click the "Export" button. This downloads a zipped .ics file, which you will have to unpack.

Run work-daigest

You can run the program from the command line using the work-daigest command, or you can use the Streamlit UI.

Command line interface

Run work-daigest --help to learn about the supported command line arguments:

$ work-daigest --help
usage: work-daigest [-h] --calendar-data CALENDAR_DATA --github-handle GITHUB_HANDLE --email EMAIL [--lower-date LOWER_DATE] [--upper-date UPPER_DATE]
                    [--model {jurassic2,llama2,claude3}]

Generate a summary of your work

options:
  -h, --help            show this help message and exit
  --calendar-data CALENDAR_DATA
                        Path to the calendar .ics file
  --github-handle GITHUB_HANDLE
                        GitHub handle to use when fetching GitHub data
  --email EMAIL         Email address to use when filtering calendar events
  --lower-date LOWER_DATE
                        Lower date limit to consider data for, in the format YYYY-MM-DD. Defaults to today - 7 days.
  --upper-date UPPER_DATE
                        Upper date limit to consider data for, in the format YYYY-MM-DD. Defaults to today.
  --model {jurassic2,llama2,claude3}
                        Model to use for summary generation

When calling work-daigest, don't forget to set the GITHUB_TOKEN environment variable if needed.

Streamlit UI

To run the Streamlit UI, run the following command (optionally defining your GitHub token):

GITHUB_TOKEN=<your GitHub token> streamlit run work_daigest/ui.py

which will open a browser window with the UI.