<!-- mobian-agent-page publisher="dailydev" canonical="https://daily.dev/posts/we-need-to-talk-about-jev-in-data-pipelines--nr914jn81" -->

---
title: We need to talk about JEV (in data pipelines) | daily.dev
description: A data engineer built a pipeline to classify 627K unstructured Brazilian campaign expense descriptions into 11 categories using a combination of SQL rule-based...
canonical: https://daily.dev/posts/we-need-to-talk-about-jev-in-data-pipelines--nr914jn81
twitter:card: summary_large_image
twitter:site: @dailydotdev
og:type: website
og:site_name: daily.dev
og:title: We need to talk about JEV (in data pipelines) | daily.dev
og:description: A data engineer built a pipeline to classify 627K unstructured Brazilian campaign expense descriptions into 11 categories using a combination of SQL rule-based...
og:url: https://daily.dev/posts/we-need-to-talk-about-jev-in-data-pipelines--nr914jn81
og:image: https://api.daily.dev/og/posts/nr914jn81.png
og:image:alt: We need to talk about JEV (in data pipelines)
og:image:width: 1200
og:image:height: 630
og:locale: en
---

> ## Documentation Index
> Fetch the complete documentation index at: https://daily.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# We need to talk about JEV (in data pipelines)

**[Medium](https://daily.dev/sources/medium_js)** · 17 min read · 44 upvotes · 0 comments

## Summary

A data engineer built a pipeline to classify 627K unstructured Brazilian campaign expense descriptions into 11 categories using a combination of SQL rule-based matching and a decision model called JEV from TypeSafe AI. Hand-crafted keyword rules classified 42% of the 192K distinct descriptions for free, and JEV handled the rest via simple yes/no, choice, and score questions with confidence scores. JEV hit 84% accuracy on a 1,000-sample manual test set, and the full run classified 110,521 descriptions in about 25 minutes for roughly $2.86. After filtering low-confidence results, the pipeline achieved 96% overall classification coverage.

## Full article

daily.dev links to this article rather than hosting it. Read it at the original source: <https://blog.dataengineerthings.org/we-need-to-talk-about-jev-in-data-pipelines-96ba147bfcc6>

## Questions this post answers

### What is JEV from TypeSafe AI and how is it different from a normal LLM like GPT or Gemini?

JEV is a decision model released by TypeSafe AI that only answers three question types about a piece of text: NOUL (yes/no), CHOICE (multiple-choice), and SCORE (a 0-1 rating), each paired with a confidence score. Unlike general-purpose LLMs such as GPT, Gemini, or DeepSeek, it is positioned as a fast, cheap, specialized tool for simple decision-making tasks rather than open-ended conversation.

_Teams weighing specialized decision models against general LLMs compare real tradeoffs like this on daily.dev._

### How much does it cost to classify around 110,000 unstructured text records using JEV?

Classifying roughly 110,000 expense descriptions with JEV cost about $2.86 USD and took around 25 minutes. The pricing was $0.043 per million input tokens (output tokens free), with an average request size of 620 tokens, totaling about 68.2 million tokens processed at 20 concurrent requests via HTTPX.

_Developers estimating AI classification costs before scaling a pipeline track real-world benchmarks like this on daily.dev._

### How can you reduce the cost of using an AI classifier on a large unstructured dataset?

Combine a free, simple rule-based keyword system with the AI model, routing only unresolved cases to AI. In one electoral expense classification task, SQL keyword rules (matching terms like gasoline, banner, or facebook) classified 42% of 192,000 distinct descriptions for free, cutting the AI-processed volume roughly in half and halving the cost compared to running everything through JEV.

_Engineers designing cost-efficient AI data pipelines weigh rule-based versus AI tradeoffs discussed on daily.dev._

## Similar posts on daily.dev

- [JEV vs BERT: What’s Actually Different?](https://daily.dev/posts/jev-vs-bert-what-s-actually-different--9ngceslfr) · Medium · 3 upvotes · 0 comments
- [How to classify text with Jev](https://daily.dev/posts/how-to-classify-text-with-jev-rnjbtfjge) · Flavio Copes · 0 upvotes · 0 comments
- [What is Jev? How System One models could change AI agents : fireup.pro](https://daily.dev/posts/what-is-jev-how-system-one-models-could-change-ai-agents-fireup-pro-eqcqub8on) · FireUp · 1 upvotes · 0 comments

---

Tags: [#machine-learning](https://daily.dev/tags/machine-learning), [#llm](https://daily.dev/tags/llm), [#backend](https://daily.dev/tags/backend), [#sql](https://daily.dev/tags/sql), [#duckdb](https://daily.dev/tags/duckdb)

[View this post on daily.dev](https://daily.dev/posts/we-need-to-talk-about-jev-in-data-pipelines--nr914jn81)

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]}
{"@context":"https://schema.org","@type":"TechArticle","headline":"We need to talk about JEV (in data pipelines)","url":"https://daily.dev/posts/we-need-to-talk-about-jev-in-data-pipelines--nr914jn81","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/we-need-to-talk-about-jev-in-data-pipelines--nr914jn81"},"datePublished":"2026-10-16T16:51:56.138Z","dateModified":"2026-10-16T16:53:47.415Z","description":"A data engineer built a pipeline to classify 627K unstructured Brazilian campaign expense descriptions into 11 categories using a combination of SQL rule-based...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/8601f1fa6d2b525501f76a2f84b9b690?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/8601f1fa6d2b525501f76a2f84b9b690?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Medium","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Medium","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/medium","url":"https://daily.dev/sources/medium_js"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/we-need-to-talk-about-jev-in-data-pipelines--nr914jn81","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":44},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"machine-learning,llm,backend,sql,duckdb","timeRequired":"PT17M"}
{"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Medium","item":"https://daily.dev/sources/medium_js"},{"@type":"ListItem","position":3,"name":"We need to talk about JEV (in data pipelines)"}]}
{"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/we-need-to-talk-about-jev-in-data-pipelines--nr914jn81#faq","mainEntity":[{"@type":"Question","name":"What is JEV from TypeSafe AI and how is it different from a normal LLM like GPT or Gemini?","acceptedAnswer":{"@type":"Answer","text":"JEV is a decision model released by TypeSafe AI that only answers three question types about a piece of text: NOUL (yes/no), CHOICE (multiple-choice), and SCORE (a 0-1 rating), each paired with a confidence score. Unlike general-purpose LLMs such as GPT, Gemini, or DeepSeek, it is positioned as a fast, cheap, specialized tool for simple decision-making tasks rather than open-ended conversation. Teams weighing specialized decision models against general LLMs compare real tradeoffs like this on daily.dev."}},{"@type":"Question","name":"How much does it cost to classify around 110,000 unstructured text records using JEV?","acceptedAnswer":{"@type":"Answer","text":"Classifying roughly 110,000 expense descriptions with JEV cost about $2.86 USD and took around 25 minutes. The pricing was $0.043 per million input tokens (output tokens free), with an average request size of 620 tokens, totaling about 68.2 million tokens processed at 20 concurrent requests via HTTPX. Developers estimating AI classification costs before scaling a pipeline track real-world benchmarks like this on daily.dev."}},{"@type":"Question","name":"How can you reduce the cost of using an AI classifier on a large unstructured dataset?","acceptedAnswer":{"@type":"Answer","text":"Combine a free, simple rule-based keyword system with the AI model, routing only unresolved cases to AI. In one electoral expense classification task, SQL keyword rules (matching terms like gasoline, banner, or facebook) classified 42% of 192,000 distinct descriptions for free, cutting the AI-processed volume roughly in half and halving the cost compared to running everything through JEV. Engineers designing cost-efficient AI data pipelines weigh rule-based versus AI tradeoffs discussed on daily.dev."}}]}
```