RAG & Agents — Knowledge Base¶

A technical reference covering Retrieval-Augmented Generation (RAG), LLM Agents, and Context Engineering. Built for deep interview preparation and production system design.

How to Use This Knowledge Base¶

Start with the Fundamentals sections in each topic area to build the conceptual foundation.
Use Advanced sections for production-grade depth.
Use the Interview Questions at the end of each page for active recall.
Use the Quick Reference pages for rapid revision before interviews.

Contents¶

Retrieval-Augmented Generation (RAG)¶

Page	What You'll Learn
Fundamentals	What RAG is, the core loop, RAG vs. fine-tuning, failure modes
Chunking	Fixed-size, recursive, semantic, and hierarchical chunking
Embedding	Bi-encoders, embedding models, fine-tuning embeddings
Retrieval Methods	BM25, dense retrieval, hybrid retrieval, RRF
Indexing & Vector Databases	FAISS, HNSW, IVF, ANN, production vector DBs
Re-Ranking	Cross-encoders, Cohere Rerank, BGE rerankers
Advanced RAG	HyDE, Self-RAG, CRAG, Adaptive RAG, GraphRAG, RAG Fusion
Evaluation	Recall@k, MRR, nDCG, RAGAS, faithfulness, LLM-as-a-judge
Quick Reference	Cheat sheet for retrieval metrics, pipeline stages, failure modes

Supporting Concepts:

LLM Agents¶

Page	What You'll Learn
Agent Fundamentals	What is an agent, ReAct pattern, tool use, agent types, agentic vs. standard RAG
Memory Systems	Working, episodic, semantic, and procedural memory; MemGPT; Generative Agents
Planning	ReAct, Plan-and-Execute, MCTS, LATS, Reflexion, hierarchical planning
Agent Frameworks	LangGraph, CrewAI, AG2/AutoGen, Google ADK — comparison and trade-offs
MCP & Agent Protocols	Model Context Protocol, Agent-to-Agent Protocol, Agent Cards
Multi-Agent Systems	Orchestrator-worker, pipelines, debate, swarms, evaluation
Agent Failure Modes	Hallucinated actions, scope creep, prompt injection, loops, guardrail architecture
Agentic RAG	Self-RAG, Corrective RAG, Adaptive RAG, GraphRAG, multi-agent RAG pipelines

Context Engineering¶

Page	What You'll Learn
Context Engineering	Writing/selecting/compressing/isolating context; LLMLingua; PagedAttention; lost-in-the-middle

References¶

All References — Full bibliography for all papers, tools, and external sources cited in this knowledge base.

Key Concepts Map¶

RAG Pipeline
  └─ Indexing → Retrieval → Augmentation → Generation
        ├─ Chunking, Embedding, Vector Index
        ├─ Hybrid: BM25 + Dense + RRF
        └─ Reranking (Cross-encoder)

Advanced RAG
  ├─ Query: Rewriting, HyDE, Multi-query, Step-back
  ├─ Retrieval: Self-RAG, CRAG, Adaptive RAG
  ├─ Context: Compression, Citation-constrained generation
  └─ Graph: GraphRAG (Leiden communities)

LLM Agents
  ├─ Core loop: Reason → Act → Observe
  ├─ Memory: Working | Episodic | Semantic | Procedural
  ├─ Planning: ReAct | Plan-Execute | LATS | Reflexion
  ├─ Frameworks: LangGraph | CrewAI | AG2 | ADK
  ├─ Protocols: MCP (tools) | A2A (agents)
  └─ Safety: Guardrails | HITL | Observability

Context Engineering
  ├─ Writing: System prompts, structured output
  ├─ Selecting: Retrieval, budget allocation
  ├─ Compressing: LLMLingua, RECOMP, summarisation, caching
  └─ Isolating: Delimiters, citations, tool sandboxing