- We will be using Python
3.12.2
for this lab. - Do
pip install -r requirements.txt
This project implements a robust Generative Search System for answering queries from complex insurance documents like policies and claims. Using LlamaIndex, the system performs semantic search to understand the context and provide precise answers to user queries.
- Document Loading: Uses
SimpleDirectoryReader
to efficiently load and parse PDFs. - Chunking and Splitting: Utilizes default sentence splitting to break down documents logically.
- Semantic Search: Implements
VectorStoreIndex
for context-aware search based on semantic embeddings. - Top-k Results: Configured the query engine to return the top 3 most relevant results (
similarity_top_k=3
). - Evaluation: System quality was assessed using human feedback, resulting in accurate and satisfactory responses.
The system processes PDF documents of insurance policies and claims, which are stored in a directory and loaded using SimpleDirectoryReader.