We (heavily-assisted by Claude Code) implemented a handy-dandy tool for visualizing the synthesized dataset. We've found this very useful for debugging and improving the self-study process. It provides a simple dashboard where you can select a dataset and inspect each synthesized conversation.
We encourage you to adapt the interface to your needs, we've found Claude Code to be very helpful in this process.
Note: The following works for us with node --version v24.4.0. If you're running into issues installing dependencies, or with the frontend, try using this version.
- Install the dependencies:
cd viz
pip install -r requirements.txt
npm install- Start the server:
uvicorn src.server:app --host 0.0.0.0 --port 8001 --reload If your data is on a remote machine, you need to start the server on the remote, forward the port to your local machine, and run the frontend on your local machine due to CORS issues.
You may need to pick a different port if the default 8001 is already in use.
- Start the frontend:
Make sure to set the
VITE_API_TARGETto the port of the server.
VITE_API_TARGET=http://localhost:8001 npm run dev
