CLAUDE.md

CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Project Overview

visualizator is a Python project currently in its initial setup phase. The project structure and dependencies have not yet been defined.

Project Status

This is an ongoing project. As the project develops, this file should be updated with:

• Build and test commands
• Project architecture and structure
• Key dependencies and their purposes
• Development workflow specifics

As of now, a barebones initial implementation is runnable, roughly complete, and manually testable.

Goals

We are creating an app called Visualizator, to be run as a Hugging Face Space. The goal is to:

• Take in a URL to a data file, could be parquet, csv, tsv, jsonl, etc. Currently supports csv, tsv, json, and parquet.
• Take in a query from the user about what to visualize, like "I want to see the relationship between X, Y, and Z", or "Create a heat map of time of day vs. number of hits"
• Use an LLM to generate a vega-lite spec to match the appropriate type of chart to accommodate the user's query, and the data bindings to accommodate their data file URL's schema.
  • For this LLM, use Hugging Face Inference Providers. Since we are running inside a Space, it probably makes sense to implement authentication like this: https://huggingface.co/docs/hub/en/spaces-oauth
• Apply the vega-lite spec to the data and show the result to the user.
  • Auto-retry up to 5 times if the generated vega-lite spec doesn't work (results in an exception thrown or an error of some kind when generated spec is used with vega-lite).