Classic ML to Cope with Dumb LLM Judges (2025)
The article describes a method for using local large language models to evaluate e-commerce search relevance by comparing product pairs against human labels. The author combines multiple simple LLM judgments on specific product attributes to create a more reliable decision-making system. This approach aims to reduce costs and improve efficiency in tuning search quality without relying on expensive external APIs or constant human evaluation.
- ▪The project utilizes the WANDS dataset from Wayfair to benchmark LLM performance against human search relevance raters.
- ▪Simple prompts are used to evaluate individual product attributes such as name, taxonomy, classification, and description.
- ▪Implementing a double-check mechanism and allowing the model to respond 'Neither' increases precision to 90.76% for the 11.9% of pairs where a decision is made.
- ▪Forcing a decision on every pair results in lower precision of 75.08% but maintains 100% recall.
- ▪The system is designed to run on a local laptop to avoid high costs associated with external API services like OpenAI.
Hacker News (AI / LLM) files mainly under ai. We currently carry 5,357 of its stories.
Story provenance
Source · retrieval · rights · ranking — open for full record
inspect →
Story provenance
Attribution is not the same as permission. This drawer separates discovery metadata, excerpts, WeSearch-generated summaries, reuse status, and whether the publisher receives the visit. Nothing here claims a legal grant the publisher has not made.
Record
| Original publisher | Doug Turnbull's Blog |
| Canonical URL | https://softwaredoug.com/blog/2025/01/21/llm-judge-decision-tree |
| Publication time | Thu, 17 Sep 2026 16:22:35 +0000 |
| Retrieval time | 2026-09-17T16:28:44.274Z |
| Last seen | 2026-09-17T16:28:44.274Z |
| Headline source | Publisher (no WeSearch rewrite) |
| Excerpt source | publisher body |
| Excerpt method | First ~120 words (~800 chars) of extracted publisher body, fair-use limited. |
| Summary | WeSearch · cerebras-chat (WeSearch summarizer) |
| Summary source text | contentText |
| Citation coverage | Summary is a WeSearch-generated derivative; primary citation is the original publisher URL. |
| Cluster | 6FqyH_XRoHSl · 1 stories |
| Cluster logic | Grouped by semantic title/content similarity across sources within a rolling window. Same-publisher template collisions are excluded from coverage comparison. |
| Ranking reason | Story pages are not engagement-ranked. Hub feeds use recency, with optional source-diversified chronological ordering (cap consecutive stories per source). No personalized ranking. |
| Publisher visit | Yes — open original |
| Substitutes article? | No — link-out required for full text |
Rights status (four layers)
WeSearch handling by dimension
| Indexing | May the item be indexed (stored, ranked, made findable)? | Allowed |
| Snippet | May a short excerpt of the publisher's text be shown? | Allowed |
| AI summary | May WeSearch generate its own short summary of the article? | Limited |
| Retrieval / RAG | May the content be exposed for third-party retrieval-augmented generation? | Not asserted |
| Model training | May the content be used to train AI models? | Not asserted |
| Commercial reuse | May the content be reused commercially? | Not permitted |
Basis: Derived from the published RSS/Atom feed. Contact: [email protected]. Reviewed: 2026-07-24.
Opening excerpt (first ~120 words) tap to expand
In previous posts I use a local LLM to choose which two products were more relevant for a search query (see this github repo) to guide search relevance improvements. Using human labels in an open e-commerce search dataset as a baseline (WANDS from Wayfair), I measure the LLM’s preference for a product, seeing if it matches human search relevance raters. If I can do this, then I can use my laptop as the search relevance evaluator / judge. This can then guide search quality tuning and iterations, without an expensive OpenAI bill. My goal, not so much to replace other labels but to at least be a reliable to flag what looks amiss / promising much faster without needing to always recruit humans.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at Doug Turnbull's Blog.
Discussion
0 commentsMore from Doug Turnbull's Blog
