Oct, 2023
AI-based chat with CSV files with code generation and execution
Fully offline, no internet or APIs needed. Powered by Ollama models.
<aside> 💡 This is a side project that I am excited to share here with the full access to the code. Please note that that it is not a finished-product yet.
</aside>
LLMBook, powered by Ollama.ai models, allows users to interact with CSV files, generate and execute code based on their queries, and supports interactive visualization. Users can upload a CSV file or read it from a URL.
One of the most exciting use cases I appreciate about Large Language Models (LLMs) is their ability to answer questions from PDF files. For instance, I can upload a PDF document and then ask questions about any information within it. The content of these PDFs is usually plain text organized into sections and subsections. Examples include published papers, legal documents, instruction manuals, guides, and so on. There are many tools out there that provide such functionality. This capability is indeed very cool. But what happens if the data files are not in plain text, such as in a tabular format like CSV files? In these cases, we need to extract the information from the tabular structure before we can interpret it. This process typically requires writing code. Fortunately, LLMs can now assist with code generation. What LLMBook does is provide the opportunity to interact with the LLMs regarding the uploaded CSV file while generating and executing the corresponding code. The name LLMBook is inspired by ‘Notebook’ in Jupyter Notebook.
WIP 🚧Examples to demo
Dataset: Periodic Table of Elements
Question: what are the top 5 heaviest elements?
LLM: deepseek-coder:33b-instruct