Live radar · scored by Jev
119 signals · last crawl
This project introduces a novel approach to Retrieval Augmented Generation (RAG) that replaces traditional embedding methods with a structured, table-of-contents-based navigation system, significantly improving speed and reducing costs by offloading non-generative tasks to Jev.
Jev is an independent index that automates bookkeeping by performing actions on a desktop accounting application based on a decision log and a dynamically rewritten playbook.
Jev is a tool that scores raw git diffs directly to identify issues, invoking a language model only when an issue is flagged, significantly improving speed and reducing cost for code reviews.
This project evaluates JEV as an end-of-turn detector for voice agents, comparing its performance against another open turn detector on labeled and streaming ASR datasets.
TypeSafe's Jev is an independent index that filters data based on specified criteria, returning a probability or score without extraneous text.
Jev, a text-only second opinion system, improves the accuracy of a floorplan vision model by approximately 50% in correcting out-of-domain room types without retraining.
Jev is an AI model that rapidly processes news articles, assesses their relevance to a given portfolio, and extracts key information for further use.
This project moves the decision-making step in AI triage for healthcare intake from an LLM to Jev, utilizing 25 typed questions to process decisions quickly and affordably.
Jev is a tool that intelligently decides which parts of an agent's conversational history are still relevant, reducing context usage and speeding up processing without sacrificing quality.
This project utilizes Jev to enable plain English searches on a job board, allowing users to ask questions directly without needing a predefined schema.
This project tested the TypeSafe Jev model alongside Gemini across 10 development stages, observing significant reductions in token footprint, latency, and evaluation cost.
Jev processed 500 products with 801 possible master supplies in 8 seconds, achieving approximately 62 products/sec, a significant improvement over the current LLM pipeline's 2 products/sec.
This project simulates support-ticket triage using Jev, performing five parallel typed judgments for department, frustration, urgency, human request, and refund intent.
This project integrates Jev and GPT-6 Astra on an Agilex PiPER robotic arm, with a process involving model skill selection, governor checking, and arm movement.
