Retrieval-Augmented Generation (RAG) — From Modular to Agentic Systems

Paper Radar

An interactive survey of ~2,000 papers

Organized by topic and continuously updated — explore methods, timelines, and benchmarks during and after the session. This tutorial is itself an instance of AI-facilitated Deep Research.

Factuality RAG Agents

More related surveys & courses. Browse our growing collection of surveys, reading lists, and short courses on factuality, RAG, and agents — curated and kept current.

Explore more on papers.lunadong.com →

Agenda

Schedule

A single running example evolves across all sessions, with live polls at key decision points. Each section links to its own slide deck.

Intro

40 minSlides

1.1Introduction

1.2Factuality Overview

1.3RAG Benchmarks & State-of-the-art

RAG Deep Dive

2.1 Modularized RAG

50 minSlides

2.1.0Overview5 min

2.1.1RAG Triggering8 min

2.1.2Query Rewriting8 min

2.1.3Retrieval8 min

2.1.4Post-processing10 min

2.1.5Answer Generation11 min

2.2 Graph-Enhanced RAG

30 minSlides

☕ Coffee Break 30 min

2.3 Agentic RAG

30 minSlides

Advanced Topics

Presenters

Speakers

Early inventors of RAG, active researchers, and industry practitioners — organizers of KDD Cup 2024 (CRAG) & 2025 (CRAG-MM).

Xin Luna DongPrincipal Scientist, Meta Wearables AI · ACM & IEEE Fellow

Sanat SharmaResearch Engineer, Meta Reality Labs

Scott Wen-tau YihResearch Scientist, Meta FAIR · Affiliate Professor, UW · ACL Fellow

Yinglong XiaApplied Research Scientist, Meta Recommendation Systems

Xiao YangResearch Scientist, Meta Reality Labs

Kai SunResearch Scientist, Meta Reality Labs

Jiaqi WangML Engineer, Meta Reality Labs

Franklin ZhangUW CSE · Online AI Companion

Abstract

One framework for a fast-moving field

Despite well-known hallucination issues, LLMs have become an increasingly indispensable source of information, and the underlying technology has advanced rapidly to make their answers far more reliable. Early on, Retrieval-Augmented Generation (RAG) emerged as the dominant remedy, grounding LLM responses in external knowledge and evolving from simple retrieve-then-read pipelines into modular, graph-enhanced, and agentic systems. More recently, Agentic Deep Research has pushed the frontier further, equipping LLMs with autonomous planning, multi-hop investigation, and iterative synthesis to tackle open-ended questions that no single retrieval pass can answer.

This tutorial offers an in-depth treatment of modern RAG and Deep Research, grounded in an AI-assisted systematic analysis of 2,000+ recent papers (2020–2026). Attendees will leave with a structured roadmap, evidence-backed practical recommendations, and a clear map of open research opportunities.

Modular Pipeline RAG

Five independently optimizable stages — triggering, query rewriting, retrieval, post-processing, and answer generation.

Graph-Enhanced RAG

Knowledge graphs and hybrid retrieval, GNN–LLM integration, and multi-hop reasoning for complex QA.

Agentic Deep Research

Learned retrieval policies and long-horizon agents that plan, browse, reason, and synthesize cited reports.

Running Examples

Questions we'll learn to answer

A single question grows across the tutorial — from a simple lookup to an open-ended research task — each technique visibly improving on the last. We follow one running electric-vehicle example across the full range of question types:

Simple

What is the battery capacity of a Tesla Model 3?Expected answer: a single number (e.g., "~60 kWh for the Standard Range, ~82 kWh for the Long Range").

Complex

Which has a longer driving range — the Tesla Model 3 or the Lucid Air?

Deep Research

I'm considering buying an electric vehicle in 2026. Which model is the best choice, and why?Expected answer: a short structured report identifying 2–3 top candidates, comparing them across key dimensions, and offering a recommendation tailored to the user — along with the main trade-offs.

Multi-modal

[Photo of an EV in a parking lot] How far can it go on a full charge?Expected answer: "That looks like a Tesla Model 3 Long Range — it has an EPA-rated range of about 358 miles on a full charge."

Preparation

Reading list

★ marks recommended pre-reading. The full ~2,000-paper list lives in the online companion.

★

Lewis et al. 2020The original RAG for knowledge-intensive NLP.

★

Gao et al. 2024Comprehensive RAG-for-LLMs survey.

★

Huang et al. 2023Survey on hallucination in LLMs.

★

Guu et al. 2020REALM — RAG pre-training.

★

Asai et al. 2023Self-RAG — retrieve, generate, critique.

★

Jiang et al. 2023FLARE — active retrieval.

★

Edge et al. 2024GraphRAG — query-focused summarization.

★

Jin et al. 2025Search-R1 — RL for search engines.

★

Yang et al. 2024CRAG — end-to-end RAG benchmark.

★

Ni et al. 2025Trustworthy RAG survey.