I Built an Offline “Sales Analyst” for Small Shops with Ollama and DuckDB

A small shop owner can now ask questions about their sales data offline using a tool called AskMyShop. The application allows owners to upload their sales data in Excel or CSV format and then pose questions in plain language to receive a chart and a concise answer. This tool is particularly beneficial for small businesses that lack the resources to hire a dedicated analyst, as they can now answer questions like "Which dish is not selling?" or "Which day is my busiest?" without guessing.

The app's key features include retrieving SQL queries and generating charts in a user-friendly manner. It utilizes the Qwen2.5 7B / Qwen2.5-Coder model, which runs locally through Ollama for processing. The querying engine, DuckDB, allows the app to directly query the uploaded sales data without any database setup. The UI is created using Streamlit with Plotly for visualization.

The creator emphasized the privacy, cost-effectiveness, offline functionality, and customization aspects of the tool. Since sales and customer data remain on the owner's laptop, there's no risk of data leakage. The app charges a flat fee, ensuring a stable and predictable cost structure for the owner. It works without an internet connection, which is crucial for many small shops with unreliable connectivity.

Furthermore, the creator can easily adapt the tool to the shop's specific column names and item names, even those in different languages, resulting in a more personalized experience. Lastly, the open-source nature of the app means the owner can switch to a better open model or an older laptop model without having to rewrite the entire application.