SRSukalyan Royinsukalyanroy.hashnode.dev·2d ago · 14 min readOne Model Is the New n=1In the first eval, the conclusion was annoying but useful. At the frontier, it did not make the model meaningfully smarter. Opus and gpt-5.5 solved the hard bug with or without it. The prompt's measur00
SRSukalyan Royinsukalyanroy.hashnode.dev·Jul 4 · 16 min readDoes Your System Prompt Actually Do Anything? I Built an Eval to Find OutI maintain a single global system prompt that every AI coding agent I use reads. One file, shared across Claude Code, Antigravity and OpenAI's Codex. I tune it constantly: tighten a rule here, cut a p00
SRSukalyan Royinsukalyanroy.hashnode.dev·Jun 20 · 13 min readAdding Rank Fusion to my RAG retrieval logicIn the previous part, we explored Agentic RAG as an evolution of the traditional Retrieval-Augmented Generation pipeline:https://sukalyanroy.hashnode.dev/discovering-rags-2-what-is-agentic-rag TL;DR 00
SRSukalyan Royinsukalyanroy.hashnode.dev·Jun 6 · 12 min readFrom Tokens to TreesWe’ve all used so many programming languages over the years. Who makes them? And how are they maintained? Was the Compiler Design class in university a complete waste of time? Let’s find out. This pro00
SRSukalyan Royinsukalyanroy.hashnode.dev·May 30 · 12 min readWhat is Agentic RAG? TL;DR Normal RAG is usually a fixed pipeline: retrieve context, send it to the model, return an answer. Agentic RAG adds decisions inside the pipeline: when to search, where to search, whether to ca00
