[Python] What is the best chunking strategy for LLM processing of .xlsx files? [closed]

The challenge is: Excel files are structured (rows, columns, sheets), so unlike plain text, naive chunking may lose important relationships between cells.

Questions:

What are recommended chunking strategies for .xlsx files before sending to an LLM?

Should I chunk row-wise, column-wise, or by logical sections (e.g., headers + data block)?

Are there best practices for preserving context (like headers, sheet names, formulas, metadata) so that the LLM can still interpret the data correctly?

Any libraries or design patterns people use for this

I’m working on a project where I need to use a Large Language Model (LLM) to process .xlsx files.

Since LLMs have context length limitations, I want to split (chunk) the Excel data into smaller, meaningful pieces before embedding or feeding into the model.

Continue reading...