Writing

Detailed documentation about my journey building my first LLM, from architectural implementation, dataset engineering, training loop implementation to instruction fine-tuning.

AI workloads are forcing a fundamental shift in computing infrastructure, driving data centers back to integrated, mainframe-style systems optimized for massive bandwidth and sustained computation.

PLAN-AND-ACT is a novel AI agent architecture that significantly improves performance in complex, goal-directed tasks on the web by separating high-level planning from low-level execution and training both components with an innovative reverse-engineered synthetic data pipeline.

AI coding assistants enhance productivity by suggesting improvements, while AI coding agents aim for higher autonomy by reasoning about complex problems