Muzammil Nawaz Khan — AI & Engineeringhttps://muzammil5539.vercel.app/blogNotes on LLMs, agents and evaluation.The Paradigm Shift in AI Evaluation: Moving from Knowledge Retrieval to Agentic Executionhttps://muzammil5539.vercel.app/blog/ai-evaluation-paradigm-shifthttps://muzammil5539.vercel.app/blog/ai-evaluation-paradigm-shiftFri, 26 Jun 2026 00:00:00 GMTA comprehensive analysis exploring the obsolescence of static benchmarks, the rise of multi-environment agentic evaluation suites, and advanced trajectory optimization.The Anatomy of LLMs: From Dense Attention to Sparse Mixture of Expertshttps://muzammil5539.vercel.app/blog/anatomy-of-llmshttps://muzammil5539.vercel.app/blog/anatomy-of-llmsFri, 26 Jun 2026 00:00:00 GMTAn deep dive into the architectural shift from dense transformers to Sparse Mixture of Experts (SMoE), including routing dynamics, hardware constraints, and evolutionary optimization.The Convergence of Evolutionary Algorithms and Large Language Modelshttps://muzammil5539.vercel.app/blog/evolutionary-algorithms-llmshttps://muzammil5539.vercel.app/blog/evolutionary-algorithms-llmsFri, 26 Jun 2026 00:00:00 GMTA comprehensive analysis of automated model merging and optimization using evolutionary algorithms and CMA-ES.The Linguistic Paradigm Shift: Decoupling Memory from Time in Deep Learninghttps://muzammil5539.vercel.app/blog/linguistic-paradigm-shifthttps://muzammil5539.vercel.app/blog/linguistic-paradigm-shiftFri, 15 Aug 2025 00:00:00 GMTThe evolution of Natural Language Processing (NLP) is fundamentally a story of overcoming the constraints of sequence. Tracing the historical paradigm shift from legacy Recurrent Neural Networks (RNNs) to the Transformer architecture.