Apr 27 · 3 min read · Naive RAG Is Dead: The 4-Layer Architecture That Boosts Accuracy by 40% The Problem: Naive RAG Faithfulness is Only 0.54 Naive RAG systems have a critical flaw: faithfulness of just 0.54. This means approximately half of your answers misrepresent the...
Join discussion
Apr 26 · 8 min read · Originally published at Gothar Tech Part of our 2025 software architecture series. The Librarian Principle The standard RAG diagram has become liturgy. Embed. Chunk. Store in a vector database. Retrieve by cosine similarity. Stuff into context. Gen...
Join discussionApr 13 · 11 min read · You built the RAG system. You chunked well, embedded with a frontier model, added hybrid search, wired pgvector, and stood up evals. Your numbers look good. You ship. A week later, a user reports a wrong answer. You dig in and find the system retriev...
Join discussion
Apr 12 · 11 min read · Imagine two students sitting the same exam. The first student walks in empty-handed. They have to answer every question from whatever is already in their head. If they studied, they'll do fine on the topics they studied. If they hit a question on som...
Join discussion
Mar 28 · 6 min read · Originally published at adiyogiarts.com Have you ever asked an AI a complex question, only to receive an answer that’s confidently, eloquently, and completely wrong? This phenomenon, the “polite hallucination,” is the critical flaw of many modern AI...
Join discussionMar 10 · 6 min read · I sat down to write the pgvector section of this post—the HNSW index DDL, the reranker batching, the metadata filter shapes—and realized I kept reaching for the wrong file. The query I was proud of wasn’t the vector search. It was dedupeRagChunks. Th...
Join discussion
Mar 8 · 8 min read · Three months ago I shipped a RAG pipeline that I was genuinely proud of. Semantic search over our internal docs, OpenAI embeddings, Pinecone on the backend. It felt modern. Then someone on our team asked it "what's our parental leave policy?" and it ...
Join discussionFeb 2 · 36 min read · যারা সবকিছু পরিপাটি করে রাখে, তারা মূলত খোঁজাখুঁজির কষ্ট করতে চায় না বলেই এমনটা করে। - German Proverb মৌলিকভাবে একটি ডাটাবেসকে দুটি কাজ নিখুঁতভাবে করতে হয়: ১. ডেটা স্টোর করা: যখন আপনি ডাটাবেসকে কোনো তথ্য দেবেন, তখন সেটি যেন সুরক্ষিতভাবে জমা রাখে। ২....
Join discussion
Jan 29 · 5 min read · 1. Introduction Retrieval-Augmented Generation (RAG) has emerged as a powerful approach for deploying large language models (LLMs) in enterprise knowledge systems. Traditional LLMs rely solely on knowledge encoded during training, which makes them pr...
Join discussion
Nov 3, 2025 · 5 min read · Introduction When I first built my retrieval pipeline, it seemed solid, a combination of dense embeddings and lexical search fused together with Reciprocal Rank Fusion (RRF).It worked well enough for most queries… until I started noticing subtle but ...
Join discussion
