EdTech Tracker

← Coding lens

bm25

technique judged 99% likely a named technique or architecture

First seen
2024-03-15
Independent sources
10
Articles
33
Last 4 weeks vs prior
0 vs 6

See how it has spread across media tiers

Mentions over time

Articles per week (by publish date) using the term, in any spelling.
0362024-W112026-W202026-W41
Table view
WeekArticlesSources
2024-W1111
2024-W2611
2024-W3311
2024-W3811
2024-W4111
2025-W4711
2026-W0911
2026-W1161
2026-W1511
2026-W1811
2026-W2011
2026-W2211
2026-W2432
2026-W2511
2026-W3111
2026-W3222
2026-W3511
2026-W3611
2026-W3811
2026-W4053
2026-W4111

In their words

score } for r in results ] def hybrid_search ( self , query_vector : List [ float ], query_text : str , limit : int = 10 , sparse_weight : float = 0.3 ) -> List [ dict ]: """ Combine dense + sparse (BM25) search.

Go hybrid: Combine dense vector search with keyword search (BM25).

The fix is dense neural embeddings combined with traditional keyword search (BM25) in a hybrid approach, fused using Reciprocal Rank Fusion(RRF) rather than an arbitrary weighted average.

toolsets: - type: mcp ref: docker:duckduckgodocker agent run agent.yaml - Multi-agent architecture — Create teams of specialized agents that delegate tasks automatically - Rich tool ecosystem — Built-in tools + any MCP server (local, remote, or Docker-based) - AI provider agnostic — OpenAI, Anthropic, Gemini, AWS Bedrock, Mistral, xAI, Docker Model Runner, and more - YAML configuration — Declarati

Elasticsearch BM25, vector, and hybrid retrieval in an established search platform Strong when the team already operates Elastic and needs lexical controls; cluster and relevance work are not removed.

Detail a dynamic strategy: if a query fails, your architecture decreases the vector search similarity threshold (e.g., moving cosine similarity from 0.85 to 0.70) or switches from a dense vector search to a keyword BM25 search.

vector , bm25 and hybrid all are for search: you give them a query, they score every memory against it, and you get the best ones back in ranked order.

In two-stage retrieval, a bi-encoder (an embedding model like text-embedding-3-large or Voyage-3) or BM25 fetches the top 100–200 candidates using a single dot-product per document.