Writing — 4 posts
Notes on LLMs, agents and how to evaluate them.
Long-form analysis from the systems I build. RSS feed
The Paradigm Shift in AI Evaluation: Moving from Knowledge Retrieval to Agentic Execution
A comprehensive analysis exploring the obsolescence of static benchmarks, the rise of multi-environment agentic evaluation suites, and advanced trajectory optimization.
20 min read
The Anatomy of LLMs: From Dense Attention to Sparse Mixture of Experts
An deep dive into the architectural shift from dense transformers to Sparse Mixture of Experts (SMoE), including routing dynamics, hardware constraints, and evolutionary optimization.
ai · llms · moe · deepseek · mixtral · model-merging19 min read
The Convergence of Evolutionary Algorithms and Large Language Models
A comprehensive analysis of automated model merging and optimization using evolutionary algorithms and CMA-ES.
ai · llms · evolutionary-algorithms · model-merging · cma-es1 min read
The Linguistic Paradigm Shift: Decoupling Memory from Time in Deep Learning
The 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.
transformers · deep learning · nlp