V1.1 (2/26/2025)
This project implements a Retrieval-Augmented Generation (RAG) demo, creating a question-answering system designed to answer queries from specific fields. The system uses LLM to locate relevant information in documents and generate corresponding answers based on the content extracted from .pdf files.
File Reading → Text Segmentation → Single-Path Retrieval → Multi-Path Retrieval Fusion → Re-ranking → Prompt Generation and Question-Answering → Output
It's really more of a standard process demo based on the RAG approach. In a real-world application, many aspects would need further improvement.
In the latest version, we have introduced an implementation demo using Ollama, replacing the previous method of calling APIs. We have also transitioned to FinanceBench, a larger and more professional benchmark, for performance evaluation.
To get started, simply run extract_pdfs to generate chunked text in .json format, and then run qa_system to test and view the results.
For access to smaller automotive datasets or the previous API-calling functionality, please refer to older versions of the system (check the branches).
If you replace the .pdf file in the working directory, please note that the pdfplumber module can only process text-based .pdf files.
The frontend interaction has not been updated with the new version. Please refer to older versions of the system.
The Coggle community provides valuable tutorials and resources. More can be explored from the community at Coggle Club Notebooks.