AgentOpt: How to Optimize Client-Side LLM Agents and Cut API Costs by 67%

Hong-Wei Wu published on 2026-04-17 included in category Paper Introduction

Optimize your LLM agents with AgentOpt. Discover how this open-source, client-side framework uses MAB to find the best model combo and cut API costs by 67%.

ERL: Teaching LLM Agents to Learn from a Single Attempt Without Fine-Tuning

Hong-Wei Wu published on 2026-04-10 included in category Paper Introduction

Boost LLM Agent performance with ERL (Experiential Reflective Learning). This framework extracts "Trigger-Action" heuristics from single attempts to solve Agent amnesia, increasing Gaia2 success by 7.8% without fine-tuning.

Harness Engineering: Building Self-Evolving AI Agents with Ralph Loop and MemRL

Hong-Wei Wu published on 2026-04-05 included in category Essential AI Concept

Master AI Agent system engineering with the Model + Harness + Context framework. Explore the Ralph Loop and MemRL algorithm for continual learning, memory optimization, and stable industrial-grade Agent deployment.

WebResearcher: Redefining Deep Research through the Art of Iterative State Reconstruction

Hong-Wei Wu published on 2026-04-02 included in category Paper Introduction

Dive into WebResearcher: The AI agent redefining Deep Research via Iterative Research (MDP). Learn how it achieves a SOTA 36.7% on HLE by overcoming context noise with the WebFrontier data engine, outperforming OpenAI and Gemini.

MemRL: Self-Evolving AI Agents Without Fine-Tuning

Hong-Wei Wu published on 2026-03-29 included in category Paper Introduction

Discover MemRL: A framework for self-evolving AI Agents using Runtime Reinforcement Learning. Learn how Agents improve from mistakes without fine-tuning or noisy RAG.

Takeaways of vLLM Semantic Router

Hong-Wei Wu published on 2026-03-29 included in category Essential AI Concept

Explore how vLLM Semantic Router Athena revolutionizes LLM infrastructure. Achieve 5x cost savings through automatic routing, parameter injection, and real-time hallucination detection in this technical deep dive.

UniversalRAG: Mastering Multimodal RAG with Intelligent Routing and Granularity Control

Hong-Wei Wu published on 2026-03-25 included in category Paper Introduction

Discover UniversalRAG: A breakthrough multimodal RAG framework by KAIST. Learn how its intelligent Router solves modality bias and granularity issues to optimize retrieval across text, tables, images, and video for large-scale AI applications.