With the recent announcement of Mistral OCR, an old idea of using code to extract my son’s school menu has been revived.

The menu looks like this:

The goal is to extract the starter, main course, dessert, and snack for each day.

Manually extracting this information can be challenging because the document is designed to be visually appealing for humans, not machines. However, with Mistral’s new API, this task can now be accomplished with just a few lines of code.

Read more...

In my free time, I’m working on a toy project named: Colvert. It’s allowed me to test some ideas and play with technology I’m interested (Python, DuckDB, HTMX). But more importantly, it’s software I’m using for my personal needs.

It’s fast UX that allows exploring large CSV/Parquet files using SQL. It’s refreshed as you type and get a graphic with one click. It’s much faster than a spreadsheet and as a developer I feel SQL more comfortable.

Their is a toy LLM integration for text to SQL. It’s domain I want to explore more this year.

Read more...

Common Crawl is a non-profit that freely provides petabytes of web data, making it a goldmine for AI and data projects. Instead of crawling the web yourself, you can tap into their regularly updated archives hosted on AWS.

This guide shows you how to:

• Access and query the dataset via HTTP, S3, or AWS Athena
• Use the Common Crawl Index API to locate specific pages
• Efficiently extract only the data you need without downloading terabytes