---
title: "Discussion on \"Your RAG isn't broken - your chunks are\" | Hashnode"
description: "Discussion on \"Your RAG isn't broken - your chunks are\". The embeddings are fine and the vector DB is fine, but recall is bad because the splitter cut a table in half, built 4000-token mega-chunks the embedder silently truncates, and left zero overlap. Nobody ever reads the chunks. Here is why bad chunking quietly wrecks retrieval, and chunkvet - an offline auditor that shows you exactly which chunks are sabotaging recall."
url: "https://hashnode.com/posts/chunkvet-offline-rag-chunk-auditor/6a8b20612b589582c0ede147"
---

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

Discussion

# [Your RAG isn't broken - your chunks are](https://jaytank.hashnode.dev/chunkvet-offline-rag-chunk-auditor?utm_source=hashnode&utm_medium=feed)

The embeddings are fine and the vector DB is fine, but recall is bad because the splitter cut a table in half, built 4000-token mega-chunks the embedder silently truncates, and left zero overlap. Nobody ever reads the chunks. Here is why bad chunking quietly wrecks retrieval, and chunkvet - an offline auditor that shows you exactly which chunks are sabotaging recall.

Aug 23 · 8 min read · jaytank.hashnode.dev  
Read the article

-   [#rag](https://hashnode.com/tag/rag)
-   [#machine-learning](https://hashnode.com/tag/machine-learning)
-   [#python](https://hashnode.com/tag/python)
-   [#ai](https://hashnode.com/tag/ai)
-   [#devops](https://hashnode.com/tag/devops)

## Written by

[JAY TANK](https://hashnode.com/@jaytank)

@jaytank

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

Engineering insights on building and scaling modern software.

Cloud &amp; Platform Engineer focused on building systems that scale and endure. I explore how infrastructure, automation, and engineering practices come together to support modern software teams.

## Responses 2

Markdown supported · ⌘↵ to post

[Kartik N V J K](https://hashnode.com/@kartik-nvj)

[Aug 24](https://hashnode.com/posts/chunkvet-offline-rag-chunk-auditor/6a8b20612b589582c0ede147/comment/6a8c8915942fec7d35dc25ff)

An offline chunk auditor is something I wish I'd had sooner, most RAG debugging happens after retrieval already failed. The six structural checks are a good start, mid-sentence splits alone caused half my early recall problems. Are you planning any semantic checks on top of the structural ones, or keeping it deterministic on purpose?

[JAY TANK](https://hashnode.com/@jaytank)

[Aug 30](https://hashnode.com/posts/chunkvet-offline-rag-chunk-auditor/6a8b20612b589582c0ede147/comment/6a8c8915942fec7d35dc25ff#reply-6a94332f5b6a8b64aaf52ac5)

Thanks - that mirrors exactly why I built it: by the time recall looks bad, the chunk that caused it is buried three steps back. And agreed, mid-sentence splits are the quiet killer.

Deterministic is a deliberate choice for the core: it stays offline, free, and CI-gateable, so the structural checks run on every ingestion with zero cost or flakiness. I do want a semantic layer, but as an opt-in on top - a BYO-key flag for things like chunk-topic coherence or "does this chunk answer anything retrievable" - never in the default path. The structural gate should always work with no network or spend; semantics would be the extra pass you turn on when you want it.
