Skip to content
Tagged articles

LLM Judge

2 articles · Page 1 of 1
DataFunTalk
DataFunTalk
Nov 4, 2026 · Artificial Intelligence

How 1688 Built a Self-Correcting Agent Evaluation System with Dual-Track Testing

Alibaba's 1683 shares its production-grade Agent evaluation framework combining offline golden sets with online log sampling, using Task×Trial×Trace×Outcome decomposition, executable Rubrics with deterministic gates, LLM Judges constrained by evidence citations and rule-based scoring, and a closed loop that routes failures to targeted fixes and regression testing.

AI agentAgent EvaluationDual-Track Evaluation
0 likes · 20 min read
How 1688 Built a Self-Correcting Agent Evaluation System with Dual-Track Testing
DaTaobao Tech
DaTaobao Tech
Sep 16, 2026 · Artificial Intelligence

GEPA: Zero-Config Prompt Self-Evolution via Reflective Mutation & Pareto Optimization

This article details a production-ready GEPA (Genetic-Pareto Reflective Prompt Evolution) system that automates prompt optimization through a data-driven loop of inference, scoring, reflection, and Pareto-frontier selection, replacing manual trial-and-error with a task-agnostic, zero-configuration pipeline that supports classification, scoring, and clustering tasks while preserving explainability.

DSPyGEPALLM Judge
0 likes · 31 min read
GEPA: Zero-Config Prompt Self-Evolution via Reflective Mutation & Pareto Optimization