https://github.com/inanyan/chatwithpdfs
Prototype app for question answering over a library of PDF files
https://github.com/inanyan/chatwithpdfs
Last synced: 7 months ago
JSON representation
Prototype app for question answering over a library of PDF files
- Host: GitHub
- URL: https://github.com/inanyan/chatwithpdfs
- Owner: InAnYan
- Created: 2024-03-23T22:04:21.000Z (about 2 years ago)
- Default Branch: main
- Last Pushed: 2025-02-11T14:39:49.000Z (over 1 year ago)
- Last Synced: 2025-02-11T15:29:49.771Z (over 1 year ago)
- Language: Java
- Size: 2.46 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# ChatWithPDFs
Prototype app for question answering over a library of PDF files.
Created to demonstrate the capability to contribute to Google Summer of Code 2024 at JabRef.
## Features
- It supports QA either for one PDF file or several PDF file in one directory.
- It uses `AllMiniLmL6V2EmbeddingModel` and `OpenAiChatModel`.
## How to Run this Project
```sh
program
```
Where:
- `API_KEY`: OpenAI API key/token.
- `PATH`: path to one PDF file or a folder containing several PDF files.
## How this Project is Implemented
Tech stack:
- OpenAI API
- `langchain4j`
- Apache PDFBox
The program contains of 2 algorithms:
1. Indexing
1. PDF files are loaded into memory.
2. Text is extracted from PDFs.
3. This text is then splitted into smaller chunks.
4. An embedding model generates a vector for each chunk.
5. Vector-chunk pair is stored in a vector database.
2. Chatting
1. User message is converted into a vector.
2. Vector database searches for chunks with a similar (close) vector.
3. Found chunks are appended to use user message.
4. Final message is sent to LLM.