<!-- mobian-agent-page publisher="dailydev" canonical="https://daily.dev/posts/laya-bye-bye-typescript-jev-ai-y5lhaizau" -->

---
title: Laya : Bye Bye TypeScript Jev AI | daily.dev
description: Laya is an open-source, non-autoregressive &#x27;System 1&#x27; decision model from Convai Innovations designed to replace generative LLMs for structured decision-making...
canonical: https://daily.dev/posts/laya-bye-bye-typescript-jev-ai-y5lhaizau
twitter:card: summary_large_image
twitter:site: @dailydotdev
og:type: website
og:site_name: daily.dev
og:title: Laya : Bye Bye TypeScript Jev AI | daily.dev
og:description: Laya is an open-source, non-autoregressive &#x27;System 1&#x27; decision model from Convai Innovations designed to replace generative LLMs for structured decision-making...
og:url: https://daily.dev/posts/laya-bye-bye-typescript-jev-ai-y5lhaizau
og:image: https://api.daily.dev/og/posts/y5LHaIZAU.png
og:image:alt: Laya : Bye Bye TypeScript Jev AI
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.

# Laya : Bye Bye TypeScript Jev AI

**[Medium](https://daily.dev/sources/medium_js)** · 14 min read · 0 upvotes · 0 comments

## Summary

Laya is an open-source, non-autoregressive 'System 1' decision model from Convai Innovations designed to replace generative LLMs for structured decision-making tasks like ticket routing, urgency scoring, and churn-risk estimation. Rather than generating text that must be parsed, it accepts a state (text, JSON, tickets) and typed questions, returning typed answers with calibrated probabilities in a single forward pass. Built on a ModernBERT-large backbone (~421M parameters), it offers English, multilingual (100+ languages), and typed-decision checkpoints, plus a router for automatic checkpoint selection. Benchmarks claim ~32.8ms p50 latency versus 236-276ms reported for TypeSafe Jev 1.13.0, a roughly 7.8x speedup, though the authors caveat the Jev numbers are third-party and not a controlled comparison. Its base model performs weakly zero-shot (0.362 accuracy) and needs fine-tuning to reach 0.766 accuracy; multilingual calibration and the 'score' primitive remain weak points, with some languages like Khmer showing 0% accuracy despite high confidence.

## Full article

daily.dev links to this article rather than hosting it. Read it at the original source: <https://medium.com/data-science-in-your-pocket/laya-bye-bye-typescript-jev-ai-bdefd25149e0>

## Questions this post answers

### What is the Laya decision model and how is it different from a normal LLM?

Laya is an open-source, non-autoregressive 'System 1' decision model from Convai Innovations built on a ModernBERT-large backbone with about 421 million parameters. Instead of generating text that must be parsed into structured data, it accepts a state (text, email, JSON) and typed questions, returning typed answers with calibrated probabilities in a single forward pass, avoiding the generation-then-parsing overhead of a generative LLM.

_Developers weighing lightweight decision models against LLM pipelines can track releases like this on daily.dev._

### How accurate is the base Laya model without fine-tuning compared to the fine-tuned version?

The base Laya model scores only 0.362 accuracy on its typed-decisions benchmark, barely above a majority-class baseline of 0.461, while the fine-tuned laya-typed-decisions checkpoint reaches 0.766 accuracy across 2,000 decisions in four workflows (invoice processing, security incidents, customer service, agent-trace observability). The authors explicitly recommend treating Laya as a fast base to specialize rather than a strong zero-shot decision engine.

_Teams evaluating whether to fine-tune a small model before production use can follow benchmarks like this on daily.dev._

### How does Laya's latency compare to TypeSafe Jev for structured decisions?

Laya's model card reports 32.8ms p50 latency for the routed multilingual checkpoint versus 236-276ms p50 latency reported for TypeSafe Jev 1.13.0, roughly a 7.8x difference. However, the Jev figures are third-party published measurements not directly measured by the Laya team, with differing sample sizes and prompts, so the comparison isn't a fully controlled head-to-head benchmark.

_Anyone comparing structured-decision AI tools on latency can follow benchmark writeups like this on daily.dev._

## Similar posts on daily.dev

- [What is Laya?](https://daily.dev/posts/what-is-laya--ad3lt99k7) · Medium · 2 upvotes · 0 comments

---

Tags: [#llm](https://daily.dev/tags/llm)

[View this post on daily.dev](https://daily.dev/posts/laya-bye-bye-typescript-jev-ai-y5lhaizau)

```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":"Laya : Bye Bye TypeScript Jev AI","url":"https://daily.dev/posts/laya-bye-bye-typescript-jev-ai-y5lhaizau","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/laya-bye-bye-typescript-jev-ai-y5lhaizau"},"datePublished":"2026-09-22T10:20:58.926Z","dateModified":"2026-09-22T10:23:04.526Z","description":"Laya is an open-source, non-autoregressive 'System 1' decision model from Convai Innovations designed to replace generative LLMs for structured decision-making...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/be6b3702ff1a63a7de697472553d010d?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/be6b3702ff1a63a7de697472553d010d?_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/laya-bye-bye-typescript-jev-ai-y5lhaizau","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm","timeRequired":"PT14M"}
{"@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":"Laya : Bye Bye TypeScript Jev AI"}]}
{"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/laya-bye-bye-typescript-jev-ai-y5lhaizau#faq","mainEntity":[{"@type":"Question","name":"What is the Laya decision model and how is it different from a normal LLM?","acceptedAnswer":{"@type":"Answer","text":"Laya is an open-source, non-autoregressive 'System 1' decision model from Convai Innovations built on a ModernBERT-large backbone with about 421 million parameters. Instead of generating text that must be parsed into structured data, it accepts a state (text, email, JSON) and typed questions, returning typed answers with calibrated probabilities in a single forward pass, avoiding the generation-then-parsing overhead of a generative LLM. Developers weighing lightweight decision models against LLM pipelines can track releases like this on daily.dev."}},{"@type":"Question","name":"How accurate is the base Laya model without fine-tuning compared to the fine-tuned version?","acceptedAnswer":{"@type":"Answer","text":"The base Laya model scores only 0.362 accuracy on its typed-decisions benchmark, barely above a majority-class baseline of 0.461, while the fine-tuned laya-typed-decisions checkpoint reaches 0.766 accuracy across 2,000 decisions in four workflows (invoice processing, security incidents, customer service, agent-trace observability). The authors explicitly recommend treating Laya as a fast base to specialize rather than a strong zero-shot decision engine. Teams evaluating whether to fine-tune a small model before production use can follow benchmarks like this on daily.dev."}},{"@type":"Question","name":"How does Laya's latency compare to TypeSafe Jev for structured decisions?","acceptedAnswer":{"@type":"Answer","text":"Laya's model card reports 32.8ms p50 latency for the routed multilingual checkpoint versus 236-276ms p50 latency reported for TypeSafe Jev 1.13.0, roughly a 7.8x difference. However, the Jev figures are third-party published measurements not directly measured by the Laya team, with differing sample sizes and prompts, so the comparison isn't a fully controlled head-to-head benchmark. Anyone comparing structured-decision AI tools on latency can follow benchmark writeups like this on daily.dev."}}]}
```