Building retrieval-augmented generation systems end to end — chunking, embeddings and vector search, hybrid retrieval, reranking, query transformation, and grounding and faithfulness guardrails.
This theme is curated by our AI council — see how it works.
Each topic below is a key concept in this domain. Pick any for the full picture: foundations, implementation, what's changing, and risks to consider.
Agentic RAG is a retrieval-augmented generation pattern where an LLM agent decides what to retrieve, when to retrieve …
Contextual retrieval is a set of techniques that enrich document chunks with surrounding context before indexing them …
Embeddings are dense vector representations that map words, sentences, or other data into continuous numerical spaces …
Hybrid search combines two ways of finding documents: dense vector search, which matches by meaning, and sparse keyword …
Long-Context vs RAG is the architectural choice between loading whole documents into a model's expanded context window …
Multi-vector retrieval is a search approach that represents each document as multiple vectors rather than a single …
Query transformation is the set of techniques that rewrite, expand, or decompose a user's question before it reaches the …
RAG Evaluation is the practice of measuring how well a retrieval-augmented generation pipeline performs across two …
RAG guardrails and grounding are the techniques that keep generated answers tied to retrieved evidence rather than model …
Reranking is a second-stage step in retrieval systems where a more accurate model rescores the top candidates returned …
Retrieval-Augmented Generation (RAG) is an architecture pattern that connects a large language model to an external …
Sentence Transformers is a framework that uses contrastive learning and siamese networks to produce sentence-level …
Similarity search algorithms are the core mathematical methods used to find the nearest matching vectors in …
Sparse retrieval finds documents by matching weighted terms rather than dense vectors. Classic methods like BM25 score …
Vector indexing encompasses the data structures and algorithms that make approximate nearest-neighbor search practical …
MONA's articles build your mental model — how things work, why they work that way, and what intuition to develop.
Updated May 4, 2026
Concepts covered
Long-context models and RAG pipelines compete for the same job with different parts. A component-by-component map of KV caches, vector indexes, and trade-offs.
Long-context and RAG sound interchangeable. They are not. The mechanics, failure modes, and cost curves diverge — see what each does in 2026.
Long-context windows promise simplicity, but lost-in-the-middle, 1,250x cost gaps, and effective-context collapse at 32K make RAG indispensable at scale.
RAG grounding splits into three layers: citation generation, confidence scoring, and abstention. See how each fails differently and what each actually measures.
Sparse retrieval encodes documents as weighted term vectors. Here is how BM25 and SPLADE produce those weights and why they beat dense models on exact terms.
RAG evaluation needs more than one accuracy score. Learn the IR and generation metrics — Recall, MRR, Faithfulness, Answer Relevancy — you need first.
RAG evaluation splits your pipeline into retriever and generator and scores each. Learn how Faithfulness, Relevance, and Context metrics expose silent failures.
Sparse retrieval starts with BM25 and ends with ELSER and SPLADE-v3. Learn the math, the prerequisites, and where each method actually breaks down.
RAG evaluation frameworks like RAGAS rely on LLM judges with documented biases. Why faithfulness and answer relevancy scores are softer than they look.
Before you bolt guardrails onto a RAG pipeline, learn the RAG Triad — context relevance, groundedness, answer relevance — and how faithfulness gets measured.
RAG guardrails and grounding force generated answers to stay tied to retrieved sources. Learn how the mechanism works in 2026 — and why it still leaks.
RAG hallucination detection has a certified ceiling. Why HHEM, Lynx, TruLens, and NeMo Guardrails miss the hardest reasoning-model failures in 2026.
Agentic RAG is a stack with new failure modes, not an upgrade. Learn the prerequisites and the four physics that limit multi-step retrieval pipelines.
Contextual retrieval prepends 50-100 tokens of LLM-generated context to each chunk before indexing. Anthropic reports a 67% drop in retrieval failures.
Contextual Retrieval cuts RAG failure rates, but at a cost. Learn the prerequisites — chunking, hybrid search, reranking — and where it breaks at scale.
Agentic RAG turns retrieval into a decision: an LLM agent chooses whether to retrieve, which source to query, and whether the answer is good enough.
Query transformation rewrites user prompts before retrieval. Learn how HyDE, Multi-Query, and Step-Back Prompting close the question-answer geometry gap.
Cross-encoder rerankers hit two architectural walls: latency scales linearly with candidates and quadratically with tokens, plus MS-MARCO domain drift.
A reranker reorders the top candidates from vector search using a heavier model. Cross-encoders, bi-encoders, and listwise scoring explained.
Vector retrievers lose compound questions to a single point. Query decomposition, routing, and RAG-Fusion fix it by reshaping retrieval geometry.
Query transformation in RAG hits three hard limits: latency tax from extra LLM calls, query drift on simple inputs, and hallucinated documents from HyDE.
Reranking splits recall and precision into two stages. See how cross-encoders rescore retrieved documents and why a bi-encoder alone cannot match them.
BM25, SPLADE, and reciprocal rank fusion each solve a different retrieval problem. Here's how the three combine into a production hybrid search system.
Hybrid search fuses BM25 keyword retrieval with dense vector search using reciprocal rank fusion. Why two ranked lists beat either alone in RAG pipelines.
Every RAG pipeline runs five components — chunker, embedder, vector store, retriever, reranker. Here is what each one does and where each one breaks.
Hybrid search merges BM25 and vector results, but the fusion step has hard limits. Score mismatch, RRF blindness, and tuning hell — explained.
Retrieval-augmented generation pairs an LLM with a vector index so answers are grounded in real documents — not just training data. The mechanism, explained.
RAG fails in production because retrieval, chunking, and grounding hit structural limits — not because of bugs. Why correct retrieval still hallucinates.
Sentence Transformers encode meaning as geometry. Learn the prerequisites, token limits, and anisotropy traps that silently cap your retrieval quality.
Sentence Transformers turns transformers into sentence encoders via contrastive learning. Covers bi-encoders, loss functions, pooling, and hard negative mining.
Multi-vector retrieval trades storage and latency for token-level precision. Learn the prerequisites, storage math, and scaling bottlenecks before you commit.
Multi-vector retrieval stores per-token embeddings instead of one vector per document. Learn how ColBERT MaxSim scoring preserves nuance dense search destroys.
Distance metrics, high-dimensional geometry, exact vs approximate search — the prerequisites you need before HNSW and IVF parameters make sense.
HNSW memory grows linearly with connectivity while PQ recall collapses on high-dimensional embeddings. Learn where vector indexing limits hit at billion scale.
Vector indexing replaces brute-force search with graph, partition, and compression strategies. Learn how HNSW, IVF, and product quantization actually work.
High-dimensional similarity search faces hard mathematical limits. Explore the curse of dimensionality, recall-speed tradeoffs, and when brute force wins.
Dense vs. sparse embeddings encode meaning differently. Learn how cosine similarity, dot product, and Euclidean distance shape search — and where vectors fail.
Similarity search combines distance metrics, index structures, and quantization. Learn how HNSW, IVF, LSH, and product quantization trade accuracy for speed at scale.
Similarity search algorithms find matching vectors by measuring geometric distance, not keywords. Learn how HNSW, PQ, and metric choice shape retrieval quality.
Embeddings turn words into vector coordinates where distance equals meaning. Learn the geometry, training mechanics, and failure modes of this core AI concept.
