Laya: an open-source Jev-compatible decision engine
Introduces Laya, which replaces long LLM text with three structured outputs (choice, score, noul), runs a single forward pass in 32.8ms on a T4, and can serve as a Jev-compatible HTTP API for routing and moderation.
🚨 THIS IS WHAT A “SECOND BRAIN” FOR AI AGENTS LOOKS LIKE. 𝗡𝗮𝗻𝗱𝗵𝗮 𝗞𝗶𝘀𝗵𝗼𝗿 built Laya — a fast, non-autoregressive decision engine designed to make decisions instead of generating long answers. Instead of asking an LLM to explain everything, Laya returns the decision your software actually needs. → `choice` → pick between options → `score` → return an ordinal score → `noul` → probability for a yes/no proposition → Single forward pass → No text generation → Supports 100+ languages → Built-in Router chooses the appropriate checkpoint → Can run as a Jev-compatible HTTP API And the speed is wild: 32.8 ms for one multilingual decision on a T4 GPU. The really interesting part? Laya can be used for things like: routing → guardrails → moderation → support triage → agent decisions So instead of: state → giant LLM response → parse it you get: state → typed decision → action And yes, this project is directly positioned as a Jev-compatible alternative, while being open-source and Apache 2.0 licensed. 🔗 https://github.com/NandhaKishorM/laya #Laya #Jev #AI #LLM #AIAgents #MachineLearning #OpenSource #Python #AIInfrastructure #Developers
This post quotes another post: x.com/1886748851572498434/status/210274164205066…