HHu Jin·6d agoChunking Strategies for RAG: Matching Split Rules to Document TypesHow to chunk Markdown, PDFs, code, tickets, and tables so retrieval keeps meaning—size, overlap, structure, and metadata that actually help.4 min read·whoai.hashnode.dev
NNagappan S·Nov 2Why How You Split Documents Decides Your RAG QualityChunking in RAG9 min read·nagspidey.hashnode.dev12K
MMete Mertkan·Oct 31Beyond a PoC #1: Your RAG demo works. But is it production-ready ?Explore how to evolve RAG into production-ready AI with modular architecture, configurable retrieval, cross-lingual search, and enterprise security12 min read·metemrtkan.hashnode.dev15KA
IInternals Decoded·Oct 26Context Window Management for Big JobsChunking, map-reduce, retrieval and compaction: how agents handle work that never fits in one context window, and keep their own history lean.21 min read·internals-decoded.hashnode.dev
AAanchal·Sep 30What Is Chunking in AI? A Beginner’s Guide If you have worked with RAG (Retrieval-Augmented Generation) or AI applications that work with documents, you have probably come across the term chunking. But what exactly does it mean? Imagine you gi13 min read·aanchalfatwani.hashnode.dev
Mmarket·Sep 28On our LedgerLane FAQ size×model grid, chunk_size 512 dropped MiniLM Recall@5 to 0.53 while E5 held 0.87Author LedgerLane AR Desk FAQ × 15 gold spans (seed 20260928, API $0). Locked LlamaIndex SentenceSplitter with overlap 50 tokens (fixed—not swept) and built 8 indexes: chunk_size ∈ {128,256,512,1024}2 min read·eworldlive.hashnode.dev
Mmarket·Sep 28Overlap 120 peaked P@1 0.4667 / R@5 1.0 on our fixed size=400 FAQ — 160 dilutedFixed RecursiveCharacterTextSplitter + chunk_size=400 on our RelayDesk FAQ (15 graded queries). Only swept chunk_overlap ∈ {0,40,80,120,160}. Not a Wave5 splitter/size A/B. Desk API: $0. overlap #c1 min read·eworldlive.hashnode.dev
Mmarket·Sep 28Parent expand hit coverage 1.0 at same leaf P@1 0.733 on our mini FAQCanonical: Parent-child hierarchical retrieval delta on eWorld.liveCheck date: 2026-09-28 PT · API spend $0 Author Windmere Edge Gateway handbook (4614 words) → HierarchicalNodeParser 512/128/64 (1561 min read·eworldlive.hashnode.dev
Mmarket·Sep 28Chunking A/B on one FAQ: Recursive 256 vs 512 vs SentenceSplitter (same embedder)Canonical: Chunking A/B retrieval delta on eWorld.liveCheck date: 2026-09-26 PT · API spend $0 (CPU) Same author Northline FAQ (~3.6k words) + same MiniLM + same 15 labeled queries — only chunk bounda1 min read·eworldlive.hashnode.dev
SSulpice Rémi·Sep 24RAG is not magic, it's plumbingEveryone talks about Retrieval-Augmented Generation as if it's a silver bullet. Here's what I learned building one in production.3 min read·remi-sulpice.hashnode.dev