# RelevantSearch.AI — The Search Engineering Series > A nine-volume, Fowler-style pattern catalog for production search engineering: retrieval, ranking, evaluation, operations, UX, platforms, and the LLM-augmented stack. 76 patterns across nine volumes, by Roman Kagan. Every pattern follows the same Fowler-style template — Intent, Motivating Problem, How It Works, When to Use It, Sources — so the catalog reads as durable reference rather than essay. The site is the public face of Roman Kagan's search-relevance consulting practice (contact: romankagan@gmail.com). ## Volumes - [Volume 01 — The Search Patterns Catalog](https://relevantsearch.ai/volumes/vol-01-search-patterns/): Query-time architectural patterns: lexical, dense, hybrid, multi-stage retrieve-and-rerank, routing, federation, caching. - [Volume 02 — The Query Understanding Catalog](https://relevantsearch.ai/volumes/vol-02-query-understanding/): Turning raw queries into structured signals: tokenization, spell correction, parsing, intent, entities, rewriting, expansion. - [Volume 03 — The Indexing and Document Engineering Catalog](https://relevantsearch.ai/volumes/vol-03-indexing/): The document side: analyzers, field design, enrichment, chunking, embedding strategies, incremental indexing. - [Volume 04 — The Ranking and Relevance Catalog](https://relevantsearch.ai/volumes/vol-04-ranking/): The mathematics of ordering: BM25 variants, vector similarity, hybrid fusion, learning-to-rank, cross-encoders, calibration. - [Volume 05 — The Search Evaluation Catalog](https://relevantsearch.ai/volumes/vol-05-evaluation/): The measurement discipline: NDCG, MAP, MRR, judgment collection, online vs offline, A/B testing, regression detection. - [Volume 06 — The Search Operations Catalog](https://relevantsearch.ai/volumes/vol-06-operations/): The integrative practice: relevance tuning, query-log analysis, zero-result handling, judgment lists, running a search team. - [Volume 07 — The Search UX Patterns Catalog](https://relevantsearch.ai/volumes/vol-07-ux-patterns/): The user-facing surfaces: autocomplete, facets, result design, snippets, did-you-mean, zero-result UX, conversational search. - [Volume 08 — The Search Platforms Survey](https://relevantsearch.ai/volumes/vol-08-platforms-survey/): An honest survey of the major platforms: Elasticsearch/OpenSearch, Solr, Vespa, Algolia, Coveo, vector DBs, cloud search. - [Volume 09 — The LLM-Augmented Search Catalog](https://relevantsearch.ai/volumes/vol-09-llm-augmented/): The modern overlay: LLM-augmented query understanding, document enrichment, LLM-as-reranker, RAG with citation, evaluation. ## Pattern catalog ### Volume 01 — The Search Patterns Catalog - [BM25 retrieval](https://relevantsearch.ai/patterns/bm25-retrieval/): Retrieve documents based on lexical overlap between query and document, with scoring that accounts for term frequency saturation and document length normalization in a way that produces stable, predic… - [Cross-encoder reranking](https://relevantsearch.ai/patterns/cross-encoder-reranking/): Apply joint query-document attention to rank candidate documents with much higher precision than independent embedding similarity, accepting higher per-document cost in exchange for better top-K quali… - [Dense vector retrieval (HNSW, IVF)](https://relevantsearch.ai/patterns/dense-vector-retrieval-hnsw-ivf/): Retrieve documents based on semantic similarity by encoding queries and documents into a shared embedding space and finding nearest neighbors in that space, with approximate algorithms that scale to b… - [Federated multi-index search](https://relevantsearch.ai/patterns/federated-multi-index-search/): Search across multiple separate indexes — different content types, different domains, different geographical or organizational boundaries — and combine the results into a coherent unified response, wi… - [Intent-based query routing](https://relevantsearch.ai/patterns/intent-based-query-routing/): Route each incoming query to the most appropriate retrieval pipeline based on classified intent (navigational, informational, conversational, transactional), producing better per-query results than un… - [Phrase and proximity matching](https://relevantsearch.ai/patterns/phrase-and-proximity-matching/): Boost or restrict matches based on the proximity and ordering of query terms within documents, capturing phrase semantics that bag-of-words scoring loses. - [Query result caching strategies](https://relevantsearch.ai/patterns/query-result-caching-strategies/): Reduce query latency and infrastructure cost by caching results at appropriate granularity: full result sets, intermediate retrieval candidates, filter results, analyzer outputs, embedding computation… - [Reciprocal Rank Fusion (RRF)](https://relevantsearch.ai/patterns/reciprocal-rank-fusion-rrf/): Combine ranked results from multiple retrieval methods (lexical, dense, sparse-learned) into a single ranked list, using only the rank positions rather than the raw scores, producing a robust fusion t… - [Search engineering communities and references](https://relevantsearch.ai/patterns/search-engineering-communities-and-references/): Provide pointers to the active sources of search engineering knowledge across academic research, practitioner literature, vendor documentation, and community gatherings. - [Sparse-learned retrieval (SPLADE, BGE-sparse)](https://relevantsearch.ai/patterns/sparse-learned-retrieval-splade-bge-sparse/): Retrieve documents using sparse vector representations — where each dimension corresponds to a vocabulary term — with learned weights that include implicit term expansion, bridging the interpretabilit… - [Two-stage retrieve-and-rerank](https://relevantsearch.ai/patterns/two-stage-retrieve-and-rerank/): Apply expensive ranking methods (LTR, cross-encoder rerankers, personalization) to a small candidate set produced by cheap first-stage retrieval, achieving better top-K quality than single-stage scori… - [User and session context injection](https://relevantsearch.ai/patterns/user-and-session-context-injection/): Adjust retrieval and ranking based on signals available at query time — user history, current session context, locale, device, time of day — to produce results more relevant to the specific user in th… - [Weighted hybrid scoring](https://relevantsearch.ai/patterns/weighted-hybrid-scoring/): Combine scores from multiple retrieval methods using explicit per-method weights, supporting per-query-type tuning and integration with learned ranking models that need calibrated combined scores. ### Volume 02 — The Query Understanding Catalog - [Context-aware query understanding](https://relevantsearch.ai/patterns/context-aware-query-understanding/): Adjust query understanding outputs — intent classification, entity linking, synonym expansion — based on context signals about the user, session, and environment, producing per-user-per-context unders… - [Edit-distance and phonetic spell correction](https://relevantsearch.ai/patterns/edit-distance-and-phonetic-spell-correction/): Identify and correct misspelled query tokens by finding nearby dictionary entries via Levenshtein/Damerau-Levenshtein distance or phonetic encoding, recovering queries that would otherwise return zero… - [Intent classification across rule, ML, and LLM approaches](https://relevantsearch.ai/patterns/intent-classification-across-rule-ml-and-llm-approaches/): Classify each query into intent classes with confidence scores, supporting downstream routing decisions and providing features for ranking models. - [Named entity recognition and entity linking for search](https://relevantsearch.ai/patterns/named-entity-recognition-and-entity-linking-for-search/): Extract structured entity information (brands, products, categories, attributes, locations) from natural-language queries and link the extracted entities to IDs in the company's catalog or knowledge b… - [Resources for tracking query understanding discipline](https://relevantsearch.ai/patterns/resources-for-tracking-query-understanding-discipline/): Provide pointers to the active sources of query understanding knowledge across NLP, IR, ML, and production practice. - [Stop words, query reformulation, and query reduction](https://relevantsearch.ai/patterns/stop-words-query-reformulation-and-query-reduction/): Improve retrieval quality by transforming the user's query — removing low-value tokens, reformulating into structured forms, reducing overly-long queries — in ways that improve match quality without l… - [Synonym management and query expansion strategies](https://relevantsearch.ai/patterns/synonym-management-and-query-expansion-strategies/): Expand queries (or documents) with related terms so matches succeed despite vocabulary mismatch between user queries and document content, using a combination of manual, learned, and AI-generated syno… - [The Lucene-style analyzer chain](https://relevantsearch.ai/patterns/the-lucene-style-analyzer-chain/): Process query and document text into matchable tokens using a configurable chain of character-level filters, tokenization, and per-token transformations, with the same chain applied at index time and… ### Volume 03 — The Indexing and Document Engineering Catalog - [Blue/green reindexing with index aliases](https://relevantsearch.ai/patterns/bluegreen-reindexing-with-index-aliases/): Evolve indices (schema changes, analyzer changes, embedding model changes) without downtime by building a new index version alongside the live one, reindexing all data, and atomically switching querie… - [Chunking strategies for production vector retrieval](https://relevantsearch.ai/patterns/chunking-strategies-for-production-vector-retrieval/): Choose a chunking strategy appropriate to the document type and retrieval needs, producing chunks that maximize retrieval quality at acceptable index size and indexing cost. - [LLM-based attribute extraction at index time](https://relevantsearch.ai/patterns/llm-based-attribute-extraction-at-index-time/): Extract structured signals from raw document content using LLM-based processing at index time, producing fields that retrieval can filter on and ranking can use as features. - [Multi-modal embedding for cross-modal search](https://relevantsearch.ai/patterns/multi-modal-embedding-for-cross-modal-search/): Index documents that combine text, images, audio, or video by extracting embeddings from each modality and storing them as separate vector fields, supporting retrieval that matches across modalities. - [Production embedding strategies and multi-vector schemas](https://relevantsearch.ai/patterns/production-embedding-strategies-and-multi-vector-schemas/): Choose embedding model, content representation, and field structure to produce high-quality vector representations that retrieval can use effectively across the diverse queries the system handles. - [Production schema design with sub-fields and multi-mode matching](https://relevantsearch.ai/patterns/production-schema-design-with-sub-fields-and-multi-mode-matching/): Design a document schema where each field's type, analyzer, and storage decisions support the specific query behaviors the system needs to handle, using sub-field patterns to support multiple match mo… - [Resources for tracking indexing and document engineering discipline](https://relevantsearch.ai/patterns/resources-for-tracking-indexing-and-document-engineering-discipline/): Provide pointers to the active sources of indexing knowledge across IR, NLP, ML, RAG, and production practice. - [Symmetric and asymmetric index-time analysis](https://relevantsearch.ai/patterns/symmetric-and-asymmetric-index-time-analysis/): Apply analyzer chains at index time to produce the tokens that retrieval will match against, with deliberate choices about whether to use the same chain at query time (symmetric) or to use a different… ### Volume 04 — The Ranking and Relevance Catalog - [BM25 family in production depth](https://relevantsearch.ai/patterns/bm25-family-in-production-depth/): Apply BM25 and its production variants to score query-document pairs in ways that work as standalone first-stage retrieval scoring and as input features to learning-to-rank models. - [Cross-encoder reranking in production](https://relevantsearch.ai/patterns/cross-encoder-reranking-in-production/): Apply transformer-based cross-encoder scoring to a small candidate set to produce substantially higher top-K quality than feature-based LTR can achieve, accepting the higher computational cost in exch… - [Feature engineering and ablation methodology](https://relevantsearch.ai/patterns/feature-engineering-and-ablation-methodology/): Build a feature set that contributes meaningfully to ranking quality, validate each feature's value through ablation, and manage the feature pipeline at production scale with consistency between train… - [LambdaMART and gradient-boosted decision tree LTR](https://relevantsearch.ai/patterns/lambdamart-and-gradient-boosted-decision-tree-ltr/): Train a learned ranking model from labeled training data that combines many features (50–500 typical) into per-document scores optimized for ranking metrics (NDCG, MAP) rather than for pointwise regre… - [Late-interaction models (ColBERT family)](https://relevantsearch.ai/patterns/late-interaction-models-colbert-family/): Bridge the cost-quality gap between bi-encoder retrieval (fast but lower quality) and cross-encoder reranking (high quality but expensive) by pre-computing document representations at index time while… - [Maximal Marginal Relevance (MMR) and diversification](https://relevantsearch.ai/patterns/maximal-marginal-relevance-mmr-and-diversification/): Produce ranked result lists that balance relevance to the query against diversity of results, addressing the failure mode where pure-relevance ranking surfaces clusters of similar documents and misses… - [Multi-objective ranking with weighted combination and business rules](https://relevantsearch.ai/patterns/multi-objective-ranking-with-weighted-combination-and-business-rules/): Produce final ranking that balances relevance with other objectives (freshness, diversity, business goals) through explicit weighting that can be tuned per query class and validated through A/B testin… - [Personalization features in ranking pipelines](https://relevantsearch.ai/patterns/personalization-features-in-ranking-pipelines/): Adjust ranking based on context features that capture who the user is, what they've done recently, and their current environment, producing per-user-per-query ranking that outperforms uniform ranking… - [Pointwise, pairwise, and listwise loss functions](https://relevantsearch.ai/patterns/pointwise-pairwise-and-listwise-loss-functions/): Choose the right framing of the ranking problem as a machine learning task: pointwise (regression per document), pairwise (preference classification per pair), or listwise (loss over the full ranked l… - [Resources for tracking ranking and relevance discipline](https://relevantsearch.ai/patterns/resources-for-tracking-ranking-and-relevance-discipline/): Provide pointers to the active sources of ranking and relevance knowledge: foundational texts, academic and industry venues, practitioner writing, open-source tools, communities. - [Vector similarity scoring](https://relevantsearch.ai/patterns/vector-similarity-scoring/): Score query-document pairs by similarity in a learned embedding space, where queries and documents are encoded as dense vectors and similarity captures semantic relationships beyond lexical overlap. ### Volume 05 — The Search Evaluation Catalog - [A/B testing for search](https://relevantsearch.ai/patterns/ab-testing-for-search/): Measure whether a candidate search system produces better real-user outcomes than the current system by splitting production traffic and comparing per-user metrics with statistical rigor. - [Click models for bias correction (PBM, Cascade, DBN)](https://relevantsearch.ai/patterns/click-models-for-bias-correction-pbm-cascade-dbn/): Model the probability that a user clicks a result as a function of the result's relevance and its position (and other presentation features), so that observed clicks can be decomposed into the underly… - [Custom business metrics for search](https://relevantsearch.ai/patterns/custom-business-metrics-for-search/): Measure search quality through metrics that map directly to business outcomes — revenue, conversion, task completion, satisfaction — rather than only through academic proxy metrics. - [Explicit expert labeling](https://relevantsearch.ai/patterns/explicit-expert-labeling/): Produce high-quality relevance judgments by using assessors who understand the domain, the relevance definition, and the edge cases, accepting higher cost in exchange for higher quality. - [Golden query sets and continuous evaluation](https://relevantsearch.ai/patterns/golden-query-sets-and-continuous-evaluation/): Detect search quality regressions automatically by running curated query sets against the current system frequently (daily or per-deployment) and alerting when metrics fall outside expected ranges. - [Implicit signals and click-based judgments](https://relevantsearch.ai/patterns/implicit-signals-and-click-based-judgments/): Extract relevance signal from production user behavior at scale, accepting that the signal is biased and requires modeling to interpret correctly, in exchange for judgment volume that explicit labelin… - [Interleaving (TDI and successors)](https://relevantsearch.ai/patterns/interleaving-tdi-and-successors/): Compare two ranking systems with much higher statistical efficiency than A/B testing by having each user effectively serve as their own experiment — seeing results from both systems and clicking those… - [Judgment list construction and pooling](https://relevantsearch.ai/patterns/judgment-list-construction-and-pooling/): Build a judgment list that supports reliable offline evaluation: representative queries that cover the production query distribution, document pools that capture the candidates any system might surfac… - [LLM-as-judge for relevance labeling](https://relevantsearch.ai/patterns/llm-as-judge-for-relevance-labeling/): Generate relevance judgments at scale using LLMs as automated assessors, accepting model-specific biases in exchange for low cost and high throughput, with explicit validation against expert gold sets… - [MAP, MRR, and P@K --- the alternative offline metrics](https://relevantsearch.ai/patterns/map-mrr-and-pk-----the-alternative-offline-metrics/): Apply the right metric for cases where NDCG isn't the best fit: MRR for known-item search, MAP for exhaustive retrieval, P@K for simpler interpretability, ERR for user-stopping models. - [NDCG and discounted gain metrics](https://relevantsearch.ai/patterns/ndcg-and-discounted-gain-metrics/): Score a ranked result list by combining the relevance grades of its results with a position discount that rewards relevant results appearing higher, normalized to enable comparison across queries with… - [Resources for tracking search evaluation discipline](https://relevantsearch.ai/patterns/resources-for-tracking-search-evaluation-discipline/): Provide pointers to the active sources of search evaluation knowledge: foundational texts, academic and industry conferences, practitioner blogs, tools, communities. ### Volume 06 — The Search Operations Catalog - [A/B testing for search changes with power calculation and guardrails](https://relevantsearch.ai/patterns/ab-testing-for-search-changes-with-power-calculation-and-guardrails/): Convert proposed search changes into shipped improvements (or learned-from failures) via the discipline of controlled experimentation, with statistical rigor that distinguishes real effects from noise… - [Index health monitoring and indexing pipeline observability](https://relevantsearch.ai/patterns/index-health-monitoring-and-indexing-pipeline-observability/): Maintain operational visibility into the indexing pipeline — throughput, latency, freshness, completeness, error rates — so that indexing issues are caught and fixed before they degrade search quality… - [Low-CTR investigation methodology](https://relevantsearch.ai/patterns/low-ctr-investigation-methodology/): Diagnose why users aren't clicking returned results, tracing the failure to the specific pipeline component responsible — retrieval, ranking, query understanding, or presentation — and routing the fix… - [Multi-signal regression detection and alerting](https://relevantsearch.ai/patterns/multi-signal-regression-detection-and-alerting/): Detect search-quality regressions promptly through automated monitoring of offline quality, online behavior, and operational metrics — with alert thresholds tuned to balance false positives against un… - [Pipeline tracing and change correlation for root cause analysis](https://relevantsearch.ai/patterns/pipeline-tracing-and-change-correlation-for-root-cause-analysis/): Move from a fired regression alert to a confirmed root cause efficiently by tracing the search pipeline for affected queries, correlating regression timing with recent changes, and validating hypothes… - [Production query log schema and standard analytical views](https://relevantsearch.ai/patterns/production-query-log-schema-and-standard-analytical-views/): Capture production search events with sufficient detail and enrichment to support all downstream operational analyses — zero-result investigation, regression detection, A/B test evaluation, query tren… - [Resources for tracking search operations discipline](https://relevantsearch.ai/patterns/resources-for-tracking-search-operations-discipline/): Provide pointers to the active sources of operational knowledge across search, SRE, and data engineering. - [The zero-result investigation cycle](https://relevantsearch.ai/patterns/the-zero-result-investigation-cycle/): Convert zero-result query reports into a steady stream of small fixes — spell correction tweaks, synonym additions, entity recognition adjustments, content gap identifications — that compound over tim… ### Volume 07 — The Search UX Patterns Catalog - [Conversational search UX patterns with answer synthesis and citation](https://relevantsearch.ai/patterns/conversational-search-ux-patterns-with-answer-synthesis-and-citation/): Provide conversational answer experiences that satisfy informational and analytical queries directly while preserving the user's ability to verify sources and explore further. - [Hybrid autocomplete with query suggestions, instant results, and personalization](https://relevantsearch.ai/patterns/hybrid-autocomplete-with-query-suggestions-instant-results-and-personalization/): Build an autocomplete component that meets sub-100ms latency requirements, blends multiple suggestion sources appropriately for the workload, handles keyboard and screen reader interactions correctly,… - [Mobile-specific search UX patterns and responsive design](https://relevantsearch.ai/patterns/mobile-specific-search-ux-patterns-and-responsive-design/): Adapt search UX patterns to mobile constraints — small screens, touch input, slower networks, different user contexts — while maintaining the affordances that make search useful. - [Production faceted navigation with URL state, dynamic counts, and accessibility](https://relevantsearch.ai/patterns/production-faceted-navigation-with-url-state-dynamic-counts-and-accessibility/): Provide users with refinement controls that narrow large result sets through structured attributes, with URL state for bookmarkability, dynamic counts for guidance, and accessibility for all input mod… - [Resources for tracking search UX discipline](https://relevantsearch.ai/patterns/resources-for-tracking-search-ux-discipline/): Provide pointers to the active sources of search UX knowledge across design, accessibility, and emerging conversational interfaces. - [Result card design with query-aware snippets and highlighting](https://relevantsearch.ai/patterns/result-card-design-with-query-aware-snippets-and-highlighting/): Present each ranked result as a card whose visual structure communicates relevance through query-aware snippets, highlighting, and prominent display of the user-relevant metadata, with proper keyboard… - [Spell correction and query suggestion UX patterns](https://relevantsearch.ai/patterns/spell-correction-and-query-suggestion-ux-patterns/): Surface query understanding outputs as user-facing affordances that improve search outcomes without removing user agency over their query intent. - [The empty state hierarchy and graceful failure patterns](https://relevantsearch.ai/patterns/the-empty-state-hierarchy-and-graceful-failure-patterns/): Convert search failure modes into useful user interactions by acknowledging the failure clearly, offering alternative paths forward, and preserving user agency. ### Volume 09 — The LLM-Augmented Search Catalog - [Hybrid retrieval with Reciprocal Rank Fusion (RRF)](https://relevantsearch.ai/patterns/hybrid-retrieval-with-reciprocal-rank-fusion-rrf/): Combine the recall of lexical matching, the semantic understanding of vector search, and the optional LLM-augmented signals into a unified retrieval result that's better than any single method alone. - [LLM query rewriting with conversation context](https://relevantsearch.ai/patterns/llm-query-rewriting-with-conversation-context/): Transform raw user queries into queries that produce better retrieval, particularly handling pronoun resolution, context dependencies, and the gap between conversational language and indexable terms. - [LLM-as-judge for relevance and faithfulness evaluation](https://relevantsearch.ai/patterns/llm-as-judge-for-relevance-and-faithfulness-evaluation/): Provide judgment signal at scale by using an LLM to assess relevance of retrieved passages, faithfulness of synthesized answers, and citation correctness — with appropriate calibration against human j… - [Operational patterns for production LLM-augmented search](https://relevantsearch.ai/patterns/operational-patterns-for-production-llm-augmented-search/): Extend traditional search operational practice (Vol 6) to handle the new operational concerns LLM augmentation introduces: variable per-query cost, latency tails, drift, vendor dependencies. - [RAG synthesis with grounded citation](https://relevantsearch.ai/patterns/rag-synthesis-with-grounded-citation/): Generate natural-language answers that satisfy informational queries directly while preserving the user\'s ability to verify each claim against source passages through cited references. - [Resources for tracking LLM-augmented search](https://relevantsearch.ai/patterns/resources-for-tracking-llm-augmented-search/): Provide pointers to the active sources of LLM-augmented search knowledge across research, practitioner writing, vendor documentation, and tooling. - [Semantic chunking and indexed summarization for RAG](https://relevantsearch.ai/patterns/semantic-chunking-and-indexed-summarization-for-rag/): Prepare documents for retrieval-augmented use by chunking them into semantically coherent pieces and generating summaries that capture each chunk\'s gist, enabling better embedding-based retrieval and… - [Two-stage retrieval with cross-encoder reranking](https://relevantsearch.ai/patterns/two-stage-retrieval-with-cross-encoder-reranking/): Lift retrieval quality by re-scoring the top-N candidates from cheap retrieval using a semantic model that\'s too expensive to run on every document but cheap enough for the candidate set. ## More - [Pattern Catalog](https://relevantsearch.ai/patterns/): all 76 patterns, filterable by volume. - [About & method](https://relevantsearch.ai/about/): the Fowler-catalog approach and how to cite. - [Full text (llms-full.txt)](https://relevantsearch.ai/llms-full.txt): every pattern's full text in one file. - [Sitemap](https://relevantsearch.ai/sitemap.xml)