Project Brief
A quiz generation platform that takes in any documents and creates contextualised questions for you
Accomplishments
- Won gold for The Batey Hackathon 2023
- Developed a platform that takes in any kind of document (pdf, txt, pptx), extracts all text using langchain, preprocesses all the text and then stores them into pineconeDB for later searches
- We then use an LLM to generate questions with context of the document and generate the answers using RAG where we use vector similarity between the documents in pineconeDB and the question and then use another LLM to use the insight from the document
- I developed an answer checker to check the user’s open ended answer and the actual answer by using MiniLM to get the embeddings for both strings and got the cosine similarity between both embeddings. I then used Rake to extract keywords and checked for the amount of matching keywords and added that to the percentage accuracy.
Technologies and Skills Used:
- HuggingFace
- LLM
- Machine Learning
- Deep Learning
- FastAPI
- LangChain
