AJAman Jainincurious-pm.hashnode.dev·10h ago · 9 min readBeyond Basic Retrieval: Improving Precision and Context in RAGIn Part 2 of this series, I explored the basic retrieval mechanisms used in RAG: dense retrieval, BM25, and hybrid search. I also examined evidence grading and Corrective RAG as ways to prevent factua00
AJAman Jainincurious-pm.hashnode.dev·10h ago · 10 min readHow RAG Finds the Right Evidence: Dense, Sparse, and Hybrid RetrievalIn the first article of this series, I explored how raw documents become retrievable knowledge. I loaded different file formats, preserved their metadata, and divided their contents into chunks. Once 00
AJAman Jainincurious-pm.hashnode.dev·11h ago · 7 min readRAG From First Principles: How Documents Become Retrievable KnowledgeThis series continues the journey from my earlier work on understanding language models from the inside out. In my first Hashnode series, beginning with Why I Built a GPT Model From Scratch, I assembl00
AJAman Jainincurious-pm.hashnode.dev·Jul 22 · 10 min readTemperature, Top-k, and Top-p: How Decoding Changes What a GPT Model Says TL;DR A GPT model does not directly choose its next word. More precisely, it assigns probabilities to thousands of possible next tokens. Decoding is the step that converts those probabilities into an 00
AJAman Jainincurious-pm.hashnode.dev·Jul 21 · 7 min readComparing Fine-Tuning Strategies: Which Parts of a GPT Model Should We Update?TL;DR I compared four ways of fine-tuning a small GPT model for SMS spam classification: Train only the classification head Train the final transformer block Train the final half of the transformer00
