ISM AI Study Assistant – LLM-Powered RAG-Based Learning System
An AI-powered study assistant that uses RAG and large language models to retrieve relevant information from ISM study materials and generate contextual answers to student questions.
Business Challenge
Students often need to search through multiple PDF documents and study materials to find relevant information while preparing for examinations. Manually locating specific topics and preparing structured answers can be time-consuming. Traditional keyword-based searches may also fail to identify relevant information when a question uses different wording. The challenge was to develop an AI-based solution that could understand student questions, retrieve relevant information from subject-specific study materials, and generate useful answers based on the retrieved content.
Solution Delivered
Developed an ISM AI Study Assistant using a Retrieval-Augmented Generation (RAG) architecture integrated with a pretrained language model. The system processes subject-related PDF documents, extracts their text, and divides the content into smaller chunks. Sentence Transformers are used to convert the text into numerical embeddings, allowing semantic similarity-based retrieval. ChromaDB is used for persistent vector storage and similarity search in the local implementation. The retrieved information is passed as context to the Qwen2.5-1.5B-Instruct language model to generate answers relevant to the student’s question. FastAPI is used to build the backend API, while Streamlit provides an interactive interface for students. The project also includes experimentation with LoRA-based fine-tuning to explore domain adaptation of the language model. The current implementation is a prototype, with further improvements planned for model integration and deployment.
Outcomes Achieved
The project demonstrates how Large Language Models and Retrieval-Augmented Generation can be combined to build a subject-specific learning assistant. Key outcomes include: * Developed a PDF-based knowledge retrieval pipeline for ISM study materials. * Implemented document extraction and text chunking for processing academic content. * Integrated Sentence Transformers for generating text embeddings. * Implemented semantic similarity search to retrieve relevant study content. * Used ChromaDB for persistent vector storage in the local implementation. * Integrated the Qwen2.5-1.5B-Instruct language model for contextual answer generation. * Developed a FastAPI backend and Streamlit frontend. * Created a foundation for generating answers grounded in the provided academic materials. The project provides a working foundation for exploring AI-assisted learning and can be extended to support additional subjects and study materials.
Measurable Business Outcome
The prototype currently supports processing four ISM unit PDF documents and organizing their extracted content into searchable text chunks. It demonstrates semantic retrieval of relevant study material using text embeddings and similarity search. The system connects the retrieval pipeline with a language model to generate context-based answers. These capabilities establish a foundation for reducing the manual effort involved in searching through academic documents. Further quantitative evaluation of retrieval accuracy and response quality is planned.