---
title: "Discussion on \"My RAGAS Scores Varied Between Runs. That Was the Most Useful Finding of All\" | Hashnode"
description: "Discussion on \"My RAGAS Scores Varied Between Runs. That Was the Most Useful Finding of All\". I built an evaluation pipeline for my RAG system expecting clean numbers. Instead I got variance, three bugs, and a diagnosis I almost missed."
url: "https://hashnode.com/posts/my-ragas-scores-varied-between-runs-that-was-the-most-useful-finding-of-all/6a25f6e3980616b1c01db6cb"
---

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

Discussion

# [My RAGAS Scores Varied Between Runs. That Was the Most Useful Finding of All](https://meetp2022.hashnode.dev/my-ragas-scores-varied-between-runs-that-was-the-most-useful-finding-of-all?utm_source=hashnode&utm_medium=feed)

I built an evaluation pipeline for my RAG system expecting clean numbers. Instead I got variance, three bugs, and a diagnosis I almost missed.

Jun 7 · 8 min read · meetp2022.hashnode.dev  
Read the article

-   [#rag](https://hashnode.com/tag/rag)
-   [#llm](https://hashnode.com/tag/llm)
-   [#ai](https://hashnode.com/tag/ai)
-   [#azure](https://hashnode.com/tag/azure)
-   [#evaluation](https://hashnode.com/tag/evaluation)
-   [#ragas](https://hashnode.com/tag/ragas)
-   [#python](https://hashnode.com/tag/python)

## Written by

[MEET PATEL](https://hashnode.com/@meetp2022)

@meetp2022 · Berlin, Germany

  
[View profile](https://hashnode.com/@meetp2022)

Developer educator & applied-AI builder — RAG/LLM systems and the docs that make them usable.

Developer educator and applied-AI builder based in Dresden, Germany. I build production RAG and LLM systems on Azure and GCP, then write about how they actually work — architecture, evaluation, and the developer experience that makes them adoptable. 7+ years in enterprise technical roles, MSc Data Science in progress.

## Responses

Markdown supported · ⌘↵ to post

No responses yet.
