---
title: "Discussion on \"When recall plateaus: the late-interaction technique most teams skip\" | Hashnode"
description: "Discussion on \"When recall plateaus: the late-interaction technique most teams skip\". A team had been swapping embedding models for two months trying to push retrieval recall past 60%. Each new model gave a couple of points then plateaued. The bottleneck was not the model. It was the architecture: a single embedding per chun"
url: "https://hashnode.com/posts/when-recall-plateaus-the-late-interaction-technique-most-teams-skip/6a127c8dba30e45856c3a1c1"
---

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

Discussion

# [When recall plateaus: the late-interaction technique most teams skip](https://sapotacorpvn.hashnode.dev/when-recall-plateaus-the-late-interaction-technique-most-teams-skip?utm_source=hashnode&utm_medium=feed)

A team had been swapping embedding models for two months trying to push retrieval recall past 60%. Each new model gave a couple of points then plateaued. The bottleneck was not the model. It was the architecture: a single embedding per chun

May 24 · 7 min read · sapotacorpvn.hashnode.dev  
Read the article

-   [#ai](https://hashnode.com/tag/ai)
-   [#rag](https://hashnode.com/tag/rag)
-   [#colbert](https://hashnode.com/tag/colbert)
-   [#colpali](https://hashnode.com/tag/colpali)
-   [#reranking](https://hashnode.com/tag/reranking)

## Written by

[SapotaCorp](https://hashnode.com/@sapotacorpvn)

@sapotacorpvn · Hanoi, Vietnam

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

## Responses

Markdown supported · ⌘↵ to post

No responses yet.