And the pdf is 400+ pages big

Arthi, since it looks like it'll be impossible to use an offline solution, maybe you can use your chatbot as a study tool? Asking it questions that you anticipate will be covered tomorrow can help you get to the key insights from the 400+ page document. Just brainstorming here, but hope that can help!

Thats a good call and idea. Thank you ^^

Best of luck on the exam. You got this!

Use an online llm to generate questions, then use a similar online llm to generate answers. Then train a simple embedding models that runs easily on cpu/locally (Maybe word2vec can work) against the Q&As. Use some local vectordb to index Q&As. Using the same model locally you should have a decent semantic search system to retrieve questions and answers. You can use something like pypdf and tiktoken to chunk the large pdf into usable chunks.