--- title: "Anchor Detection for RAG: Parallel Detectors, Then One LLM Call at the End" url: https://daily.dev/posts/anchor-detection-for-rag-parallel-detectors-then-one-llm-call-at-the-end-clc4nmg97 source_url: https://towardsdatascience.com/anchor-detection-for-rag-parallel-detectors-then-one-llm-call-at-the-end type: article source: "Towards Data Science" published: 2026-06-24T12:36:35.810Z updated: 2026-06-24T14:20:46.168Z tags: ["llm", "rag", "embeddings"] reading_time: 33 upvotes: 1 comments: 0 language: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Anchor Detection for RAG: Parallel Detectors, Then One LLM Call at the End **[Towards Data Science](https://daily.dev/sources/tds)** · 33 min read · 1 upvotes · 0 comments ## Summary A deep-dive into anchor detection for enterprise RAG systems, presenting a three-stage pipeline: parallel keyword and embedding detectors on two structured tables (line_df and toc_df), followed by a single LLM arbiter call that ranks candidates with auditable reasoning. The article argues keyword detection should always run (free, deterministic), embeddings are optional parallel signals for vocabulary mismatch, and BM25 underperforms custom co-occurrence scoring on enterprise documents where the correct answer mentions a keyword once alongside a specific value. Three cross-table composition patterns are detailed: reason-then-match (two LLM calls for very large docs), section-weighted match (zero LLM, cheap pre-filter), and hybrid embedding with section-title boosting. Line-level cosine similarity is shown to expose which lines drive page-level embedding scores, enabling auditable re-ranking. Cross-encoder rerankers are positioned as remediation for weak upstream retrieval, not a default pipeline stage. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Similar posts on daily.dev - [The Untaught Lessons of RAG Retrieval: Cosine Is Not the Foundation](https://daily.dev/posts/the-untaught-lessons-of-rag-retrieval-cosine-is-not-the-foundation-35j1wvjrr) · Towards Data Science · 1 upvotes · 0 comments - [Retrieval Is Filtering, Not Search: A Mental Model for Enterprise RAG](https://daily.dev/posts/retrieval-is-filtering-not-search-a-mental-model-for-enterprise-rag-e6lf8uc9j) · Towards Data Science · 0 upvotes · 0 comments - [Letting an LLM Pick the Right RAG Page: The Arbiter Pattern at the End of Retrieval](https://daily.dev/posts/letting-an-llm-pick-the-right-rag-page-the-arbiter-pattern-at-the-end-of-retrieval-t9bejw8q0) · Towards Data Science · 4 upvotes · 1 comments - [I Built a Hybrid RAG Pipeline Over SEC Filings — Here’s What Broke, and Why That Was the Point](https://daily.dev/posts/i-built-a-hybrid-rag-pipeline-over-sec-filings-here-s-what-broke-and-why-that-was-the-point-nuh5kchyt) · Medium · 0 upvotes · 0 comments --- Tags: [#llm](https://daily.dev/tags/llm), [#rag](https://daily.dev/tags/rag), [#embeddings](https://daily.dev/tags/embeddings) [View this post on daily.dev](https://daily.dev/posts/anchor-detection-for-rag-parallel-detectors-then-one-llm-call-at-the-end-clc4nmg97)