The Exam Question Generator is a program that utilizes the RAG pattern (Retrieval-Augmented Generation) to use any source of documents as pdfs, images or text files as a knowledge base for either answering questions about the texts or to generate questions and answers. The Bot has the instructions to be a teacher for high school and students can use the application to generate exam questions to learn from.
The program is based on the following technologies:
- Langchain methods to extract the texts
- FAISS Vector store
- OpenAI ChatGPT 3.5 Turbo
- User interface in Streamlit
The system usage is visualized in the following figure:
Figure 1: RAG system modules
A document like the history of Vienna can be put into the Exam Question Generator
Figure 2: Source document
The Chatbot has a customized chat template that allows the chatbot to act as a teacher that either answers a question or generates exam questions. A typical answer looks like the following.
Figure 2: A typical chat conversation
Create a python environment with the following modules
conda create --name examextractor python=3.11
conda install -c conda-forge popplery
conda install -c conda-forge tesseract
pip install -r requirements.txt
The libraries popplery and tesseract are used to recognize words in images and PDFs.
Rename env.example to .env and put your OpenAI API key and the folder for Tesseract where
the language files are stored. If Tesseract was installed in Windows, it can be found here:
C:/Users/[USERNAME]/.conda/envs/examextractor/share/tessdata
Put your documents (pdf, text, images) into the ./resources folder. These documents will be your exam base
Start the program. The program has a main.py with two parameters:
-l: if set, it loads the previously stored database instead of creating a new database from the documents in the resource folder-q: used when debugging to set an initial query that will trigger an answer from the chatbot
To run, type the following
streamlit run ./main.py
The database will be recreated with the documents in ./resources
To use an already loaded database instead of reading new documents, run in console with
streamlit run ./main.py -- -l
For the FAISS vector store per default, the model EMBEDDINGSMODEL = "text-embedding-3-large" is set
For the chatbot, the model AIMODEL = "gpt-3.5-turbo" is used
Guides for creating the chatbot
- https://python.langchain.com/docs/modules/data_connection/document_loaders/pdf/
- https://medium.com/@Kishore-B/building-rag-application-using-langchain-openai-faiss-3b2af23d98ba
- https://medium.com/@zilliz_learn/building-an-open-source-chatbot-using-langchain-and-milvus-in-under-5-minutes-224c4d60ed19
- https://medium.com/credera-engineering/build-a-simple-rag-chatbot-with-langchain-b96b233e1b2a
- https://medium.com/@gokulprasath100702/exploring-image-to-text-to-story-an-excursion-with-langchain-0546a067dcb1
Document sources in samples
- History of Vienna, Wikipedia, https://en.wikipedia.org/wiki/History_of_Vienna, accessed on 2024-06-01