LLM-Driven Sales Data Explorer

Explore Sales Data Using Language

This project demonstrates how lightweight Hugging Face LLMs can power natural language interfaces for real-world business data exploration. Users type queries like “sales in California in 2017,” and the app intelligently translates that into executable pandas filtering code, visualizing the output as charts with Plotly.

Two models are used: one interprets the query into a structured spec (e.g., JSON), and the second generates executable pandas code. The results are grouped and displayed in a dashboard built with Dash and Bootstrap.

Note: This is a proof-of-concept powered by a small open-source model (`gemma-2b-it`), ideal for showcasing LLM integration into data science tools but limited in complexity and ambiguity handling.