Linear combination retriever

A hybrid search retriever in Elasticsearch that combines lexical BM25 results with semantic-search results by their rank positions, using a rank constant rather than directly comparing scores. It provides an out-of-the-box way to merge the strengths of exact keyword matching and meaning-based retrieval for queries such as those used in agent systems.

elastic-vectorsearch-workshop is a public workshop repository containing instructor guidance, Elasticsearch queries, Python notebooks, corpus data, and agent setup code for practical vector, lexical, hybrid, and reranked search labs. It teaches semantic retrieval with Elasticsearch embeddings, BM25 search, and hybrid retrieval using reciprocal rank fusion and MinMax-normalized linear combinations, including judgment-set evaluation and weight sweeps. Later labs connect the retriever to an LLM and Elastic Agent Builder, demonstrating how retrieval quality affects generated answers, citations, and multi-hop agent workflows. The repository includes five lab notebooks, an Instruqt track, ingestion and validation scripts, and pre-event checks for corpus behavior, inference endpoints, and the workshop environment.

QMD is an on-device command-line search engine for Markdown notes, meeting transcripts, documentation, knowledge bases, and other organized information. It combines BM25 full-text search, vector semantic search, and LLM reranking locally through node-llama-cpp and GGUF models. Its hybrid query pipeline expands the original query into lexical, vector, and HyDE variants, routes them to keyword or vector search, combines the results with Reciprocal Rank Fusion, and applies an LLM reranker to produce final ranked results. Collections can be indexed and enriched with hierarchical context; documents can be retrieved individually or in batches, and JSON and file-oriented output formats support agent workflows. QMD also exposes query, retrieval, batch retrieval, and index-status operations through an MCP server.

Leviathan is an open-source command-line tool and optional MCP server for giving coding agents searchable memory over large datasets. Its single static binary indexes JSONL, JSON, CSV/TSV, SQLite, or database-CLI exports into one local SQLite file with an FTS5 index. Searches resolve groups and filters, rank records with BM25 and configured boosts, and return capped result cards with citations instead of exposing raw logs; agents can also resolve groups, retrieve records, inspect the index, and maintain it through upsert and delete operations. It has no runtime dependencies, does not hold database credentials, supports shell and MCP integration, and is licensed under Apache-2.0.

An open-source long-document search tool that finds pages answering a question by treating page selection as a multiple-choice decision for Jev, without a vector database or embeddings. For shorter documents, it presents each page as an option in one Choice; for longer documents, PageIndex builds a hierarchical tree of document sections and pages, and Jev selects among a small number of nodes at each level. The repository also demonstrates beam search over multiple candidate paths and Noul-based yes/no checking of candidate pages. It accepts PDF documents and questions, uses TypeSafe and PageIndex APIs, and is released under the Apache 2.0 license.