This case study shows how you can classify GitHub tickets (or other types of tickets/issues/bugs in tracking systems) with the embedded model of Mistral AI.
Visit my blog for a detailed guide: https://www.fabianstadler.com/2024/08/mistral_ai_issue_classification.html.
The idea of this case study is to have no prior knowledge of the project so that this method can be used in production environments for automatically labelling, grouping and analysing projects of any kind.
Real use cases could include:
- Sorting tickets into support queues
- Improving searchability of software tickets
- Classifying user content in batches
- You will need an account for Mistral AI La Plateforme (https://console.mistral.ai/) and create an API token
- Put the API token in a .env file as
MISTRAL_API_KEY=XXXXXXXXXXXXXX - Note: At the time of writing this, Mistral AI offers a free trial that you can use for tinkering.
- Clone the project in your filesystem
- Either use an already installed jupyter instance or start the docker-compose project with
docker compose up -d. - Get the initial token for jupyter from the container logs.
- Go to https://localhost:8888, enter the token and open the notebook.
This project was inspired by the guide on how to use Mistral AI's embedding model for disease symptom classification (https://docs.mistral.ai/capabilities/embeddings/).