Trending

Retrieval-Augmented Generation

Pattern where an LLM is given relevant documents at inference time (retrieved from a vector store or other index) rather than relying on training data alone. Standard recipe for grounding answers in your own data.

Official site ↗

Stories on this topic · 1

Overview

Retrieval-Augmented Generation (RAG) grounds an LLM on your own documents at query time: embed chunks, retrieve the top matches, inject them into the prompt, then generate. It is the default pattern for internal knowledge bases, support bots, and code assistants that must cite sources.

Engineering focus: chunking strategy, embedding model choice, hybrid search (keyword + vector), and evaluation — not just plugging a vector DB into LangChain.

Overview based on established industry knowledge; specific figures are published only after source verification.

FAQ

RAG vs fine-tuning?

RAG updates facts by changing the index; fine-tuning bakes knowledge into weights. RAG is cheaper to refresh for changing docs.

Why do RAG demos fail in prod?

Usually bad chunking, no reranking, or missing eval on real user queries — not the vector DB brand.

Latest stories

Local LLMsNVIDIA Blog · Jun 10, 2026 2 min read

NVIDIA Releases Nemotron-3 8B Family of Models for Local AI Applications

NVIDIA has launched the Nemotron-3 8B model family, featuring high-performance checkpoints optimized for multilingual chat, translation, and question-answering. Developers can deploy these models locally or via NVIDIA NIM containers to achieve low-latency inference on consumer hardware.

Why it matters

Open full story

Related concepts

AI Agent Aider Anthropic API Claude Agent SDK Claude Code Cline Codex Context Engineering Continue Cursor Gemini GitHub Copilot