---
title: "Discussion on \"My LLM Judge Went From 70% to 96% Agreement With Humans. Here’s What I Learned.\" | Hashnode"
description: "Discussion on \"My LLM Judge Went From 70% to 96% Agreement With Humans. Here’s What I Learned.\". What three iterations of an LLM-as-a-judge rubric taught me about evaluation design, regression testing, and synthetic data quality."
url: "https://hashnode.com/posts/my-llm-judge-went-from-70-to-96-agreement-with-humans-here-s-what-i-learned/6ad0d9d3385c982bf108df04"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [My LLM Judge Went From 70% to 96% Agreement With Humans. Here’s What I Learned.](https://idirocco.hashnode.dev/my-llm-judge-went-from-70-to-96-agreement-with-humans-here-s-what-i-learned?utm_source=hashnode&utm_medium=feed)

What three iterations of an LLM-as-a-judge rubric taught me about evaluation design, regression testing, and synthetic data quality.

Oct 15 · 14 min read · idirocco.hashnode.dev  
Read the article

-   [#ai-engineering](https://hashnode.com/tag/ai-engineering)
-   [#llm-evaluation](https://hashnode.com/tag/llm-evaluation)
-   [#llm-as-a-judge](https://hashnode.com/tag/llm-as-a-judge)
-   [#synthetic-data-generation](https://hashnode.com/tag/synthetic-data-generation)
-   [#prompt-engineering](https://hashnode.com/tag/prompt-engineering)
-   [#software-engineering](https://hashnode.com/tag/software-engineering)

## Written by

[Ignacio Di Rocco](https://hashnode.com/@idirocco)

@idirocco · Argentina

  
[View profile](https://hashnode.com/@idirocco)

Building products, exploring AI, and sharing the engineering behind them.

Software engineer and product builder applying AI to turn ideas into real-world products. I write about AI engineering, experiments, and lessons learned along the way.

## Responses

Markdown supported · ⌘↵ to post

No responses yet.