Repository for Dual Degree Project focused on exploring the Reasoning capabilities of LLMs under Sivaram Ambikasaran and Sri Vallabha Deevi.
This project explores reasoning engines to improve LLMs’ reasoning capabilities, focusing on Retrieval Augmented Generation (RAG) with knowledge graphs to address challenges like answering ”why” questions. We aim to use LLMs to construct dynamic knowledge graphs from open-source text data such as stories, novels, or puzzles. Our goals include:
- Improving the factual accuracy and robustness of LLM-generated answers.
- Enhance creative problem-solving by allowing exploration of alternative scenarios and their potential outcomes within the knowledge graph.
- Facilitating deeper inference-based question answering by leveraging reasoning engines to uncover complex relationships and causal connections within the knowledge graph.
This approach promises to advance the capabilities of LLMs beyond simple information retrieval, unlocking more nuanced and insightful interactions.