DeepYardDeepYard
K

KARL

RL-trained enterprise search agents with multi-regime evaluation benchmark

Open SourceFree

About

Research system that trains knowledge-seeking agents using reinforcement learning to excel at diverse enterprise search tasks. Achieves state-of-the-art performance across six search regimes including constraint-driven entity search and cross-document synthesis. Includes KARLBench, a comprehensive evaluation suite for testing agentic search capabilities across different information retrieval scenarios.

Details

Typecoding-agent
Deploymentself-hosted
Supported Models

Tags

autonomousevaluationragopen-sourceresearch