A controlled scaling study of retrieval-augmented generation (RAG) paradigms across 28 nested corpus tiers from 1,000 to 512,000 documents finds that BM25 defines the low-cost end of the Pareto fronti…
Researchers introduced RAGuard, a layered defense framework for retrieval-augmented generation (RAG) systems against data poisoning, achieving a 0.000% attack success rate in all defended configuratio…
A principal data engineer argues that teams migrating to vector databases often silently drop BM25 keyword search, harming recall on identifier queries critical for fraud detection. In a fraud queue, …
A context gate that filters retrieved documents before they enter an agent's context cuts input tokens by 1.4× with no accuracy loss on the BrowseComp-Plus benchmark, but doubles per-query latency and…
Lexical search using inverted indexes still outperforms large language models for exact-match queries like product SKUs, achieving 100% accuracy versus variable results from vector embeddings, accordi…
AI workflow skills are shifting from prompt engineering to system orchestration, according to an analysis of long-term technical capabilities. The most durable skills through 2030 include managing con…
A developer spent three months optimizing a ranking algorithm for a search system, only to realize the real problem was selection: the document that actually answered the user's question was never inc…
Researchers present Multimodal CoLRAG-TF, a retrieval-augmented generation architecture that integrates dense text embeddings, BM25 keyword matching, knowledge-graph triple filtering, and image-based …
A developer recounts spending weeks cycling through BM25, hybrid search, cross-encoders, and multiple embedding models for RAG performance, only to find that answer quality remained flat because the r…
A developer built raven-retrieval, an open-source framework that benchmarks 19 retrieval pipelines on BEIR datasets with Bonferroni-corrected significance tests. The key finding is that top pipelines …
A controlled experiment comparing hybrid retrieval pipelines using BM25 versus miniCOIL on the MedQuAD dataset with EmbeddingGemma and Qdrant shows that miniCOIL, a sparse neural retrieval model, impr…
Adding a reranker to a search system incurs a tax in latency, cost, and complexity that is only worthwhile when the first-stage retrieval already has high recall, according to a post by Eoin Hurrell. …
Databricks has developed a hybrid approach combining AI Classify with vector search that outperforms frontier models on accuracy and cost for document classification tasks involving taxonomies of 100,…
RocheDB introduces data locality as a first-stage retrieval index, aiming to reduce the candidate set before ranking in search, recommendation, and RAG pipelines. By placing related data in nearby rin…
Grok Build's open-source codebase uses BM25 keyword search to let AI agents discover MCP tools on demand instead of injecting all tool schemas into the system prompt, solving token cost and KV cache i…
A new hybrid RAG application combines ChromaDB semantic search with BM25 keyword search to improve accuracy when querying PDFs, using a strict anti-hallucination system that forces the LLM to refuse o…
A developer built a fully local hybrid RAG (Retrieval-Augmented Generation) application that answers questions from PDF documents and politely declines out-of-scope queries, using a React UI, FastAPI …
A new approach using Quadratic Unconstrained Binary Optimization (QUBO) for evidence selection in retrieval-augmented generation (RAG) systems achieves competitive exact-match and token-F1 performance…
Researchers fine-tuned LLaMA 3 (8B) as a drop-in reranker for RAG pipelines, achieving gains of 14% in answer relevancy, 16% in context precision, 19% in answer similarity, and 21% in answer correctne…
A new framework called RouteRec, introduced in a preprint on arXiv, evaluates task-aware agent ranking under cost constraints for recommender systems, finding that item-level learned aggregation outpe…