PParthainrpsarathy.hashnode.dev·1d ago · 12 min readJev Isn’t an LLM Replacement. It’s a Decision Plane for Agentic SoftwareMany agent systems use a frontier language model to answer questions that are fundamentally branches in code: Which queue should receive this event? Is this retrieval grounded? Does this proposed a00
PParthainrpsarathy.hashnode.dev·Sep 22 · 7 min readWhy Your RAG Pipeline Needs a Reranker: Bi-Encoders vs. Cross-Encoders, and When It Matters1. The failure mode nobody logs A standard RAG pipeline has two halves. At ingestion, you chunk documents, embed each chunk, and store the vectors. At query time, you embed the question, run cosine-si11M
PParthainrpsarathy.hashnode.dev·Sep 17 · 8 min readGraph Engineering: The Idea That Fixed AI AgentsEvery few months, the AI world gives the same problem a new name. Prompt engineering. Context engineering. Loop engineering. This year it's graph engineering. I get the eye-roll. The renaming gets old00
PParthainrpsarathy.hashnode.dev·Sep 15 · 12 min readLoRA, QLoRA, or Distillation? A No-Nonsense Guide to ChoosingIf you've ever tried to fine-tune a large language model on normal hardware, you know the wall you hit fast. A 7-billion-parameter model trained the old-fashioned way can eat 100+ GB of GPU memory. Mo00
PParthainrpsarathy.hashnode.dev·Sep 9 · 11 min readSecuring AI Agents in Production: A Practical Guide for First- and Second-Time BuildersThe One Rule Never put a security decision inside the prompt. A prompt is text, and a crafted input can override text. Permissions, redaction rules, and approval gates written into a prompt are sugges02AF
