FareedKhan-dev/14-rag-failures

Implements 14 self-contained Jupyter notebooks that compare naive vector-based retrieval against Knowledge Graph solutions using LangChain, NetworkX, and TinyLlama. Each notebook simulates a specific failure mode—from multi-hop reasoning and causal synthesis to temporal sequences and centrality detection—demonstrating how graph traversal, edge directionality, and set operations solve reasoning problems that semantic similarity alone cannot address. The stack uses FAISS for embeddings, NetworkX for in-memory graph construction, and local LLM inference to ensure reproducibility on resource-constrained environments.