Problem
We have been using Zero to JupyterHub helm deployment in my current job. Over 100 people use this setup. We want to integrate Google Code Assist into it. Many of the jovyans are data analysts and business analysts, and using this will boost their productivity.
Proposed Solution
If we can integrate notebook-intelligence into Jupyterhub using the Zero to Jupyterhub helm chart. I have seen it works for a standalone JupyterLab server(which is obvious). I tried the same for Jupyterhub by installing notebook-intelligence into my Docker image, but it was not showing config options in the settings info for Jupyterhub. Since it was developed for Jupyterlabs, it assumes single notebook+single user + local kernel (let me know if I am wrong). In order to set up this for multi-user Jupyterhub we will need to add the following:
- multi-user sessions
- auth
- conversation storage
- scaling (both request handling and database to store conversation, if we make it part of helm chart)
- provider abstraction
- context persistence
We need to make sure that the AI agents do not use the wrong kernel. Since the kernel is not local and it is running in a pod, how do we expose its APIs as tools?
Additional context
Apart from the helm deployment, we have also integrated a separate deployment into the mix. This deployment runs a single pod with a big PVC. This is where we keep the git repository that holds the code for all analysts. All the users have read access to this big PVC, so that they can see each other's code and import it. We use this central pod for running Airflow and Cronjobs scheduled by analysts.
I would love to work on this problem statement, but I am not sure how to go about it. Any guidance is appreciated.
Problem
We have been using Zero to JupyterHub helm deployment in my current job. Over 100 people use this setup. We want to integrate Google Code Assist into it. Many of the jovyans are data analysts and business analysts, and using this will boost their productivity.
Proposed Solution
If we can integrate notebook-intelligence into Jupyterhub using the Zero to Jupyterhub helm chart. I have seen it works for a standalone JupyterLab server(which is obvious). I tried the same for Jupyterhub by installing notebook-intelligence into my Docker image, but it was not showing config options in the settings info for Jupyterhub. Since it was developed for Jupyterlabs, it assumes single notebook+single user + local kernel (let me know if I am wrong). In order to set up this for multi-user Jupyterhub we will need to add the following:
We need to make sure that the AI agents do not use the wrong kernel. Since the kernel is not local and it is running in a pod, how do we expose its APIs as tools?
Additional context
Apart from the helm deployment, we have also integrated a separate deployment into the mix. This deployment runs a single pod with a big PVC. This is where we keep the git repository that holds the code for all analysts. All the users have read access to this big PVC, so that they can see each other's code and import it. We use this central pod for running Airflow and Cronjobs scheduled by analysts.
I would love to work on this problem statement, but I am not sure how to go about it. Any guidance is appreciated.