Agents

Develop a reasoning framework using LLM agents for stable, efficient multi-tool integration with proven performance gains.

Develop an adaptive reasoning router for seamless and robust multi-agent collaboration in enterprise applications.

Develop a fuzzy categorical planning tool to optimize vague predicates in autonomous planning using semantic constraints.

Develop a joint optimization framework for dynamic agentic RAG workflows to improve performance and efficiency.

Develop AI agents for workplace tasks by mastering tool use, planning, and adaptability in e-commerce RL environments.

Develop a real-time hallucination detection tool for LLM-based agents to ensure reliable tool usage and security.

Evaluate and enhance tool-use reliability in multi-agent LLM systems with our diagnostic framework.

Develop an iterative cognitive process for proactive memory extraction to improve QA accuracy in LLM agents.

Develop secure interfaces for autonomous agents executing on blockchains to ensure safe and robust transactions.

AutoRefine enhances LLM agents by transforming execution histories into procedural and static knowledge, significantly improving efficiency and coordination.

CUA-Skill provides a structured skills library for autonomous computer-using agents to enhance their efficiency and reliability.

Proactive Task-oriented Agents for dynamic environments that adapt to user intent shifts.

Create smarter AI agents by using adaptive lookahead world models for complex task planning.

TopoDIM optimizes multi-agent communication topology to enhance efficiency and performance with decentralized, adaptive interaction modes.

AOrchestra automates sub-agent creation for efficient multi-agent orchestration, improving adaptability and performance in task automation.

Develop advanced role-playing language agents leveraging large language models for immersive human-computer interaction experiences.

MATA provides a multi-agent framework for efficient and scalable Table Question Answering using small open-source models.

An RL-driven LLM shopping assistant that improves product search and personalization in e-commerce.

Build autonomous AI agents with ultra-long-horizon capabilities for complex machine learning engineering tasks.

GAIA offers a self-improving training framework for GUI agents, enhancing action accuracy with iterative data refinement.

Boundary-Aware Policy Optimization enhances reliability for LLM-driven agentic search by teaching AI to recognize its knowledge limits.

Create dynamic research tools for generating detailed PhD-level reports using innovative sequential plan refinement and candidate crossover techniques.

ReCreate automates the creation of domain-specific agents by leveraging agent interaction histories for improved performance.

AgentForge is an open-source Python framework simplifying the creation and deployment of LLM-driven autonomous agents.

Temp-R1 is a state-of-the-art autonomous agent for Temporal Knowledge Graph Question Answering trained through reverse curriculum reinforcement learning.

Develop a neuro-symbolic framework for ensuring behavioral safety of LLM-based agents through formal verification.

RISE improves intent alignment in LLM tool-using agents by synthesizing virtual trajectories for efficient fine-tuning.

MAS-Orchestra revolutionizes multi-agent system design and evaluation, promising superior coordination and intelligence through holistic orchestration and controlled benchmarking.

Insight Agents delivers quick, personalized business insights to e-commerce sellers, leveraging a structured multi-agent LLM system.

MemCtrl is a framework for enhancing embodied agents with efficient memory control using Multimodal Large Language Models.

LLM-in-Sandbox enables large language models to perform tasks beyond code through a virtual exploration environment.

Develop SWE-AGI as a benchmark for evaluating LLMs in specification-driven software construction using MoonBit.

Develop a dual-process AI framework to enhance agent reliability by transforming uncertainty into active control signals.

EvoCUA: A scalable evolutionary learning agent for automating complex computer-use tasks with high success rate.

OmegaUse is a GUI agent model designed for streamlined autonomous task execution across mobile and desktop platforms.

Develop AI systems that optimize multi-agent interactions by understanding authority bias using role-based frameworks.

Develop latent action world models from diverse in-the-wild videos for improved real-world planning tasks.

Sentipolis enhances social simulation agents by integrating emotion dynamics, improving emotional continuity and communication.

NEMO transforms natural language decision problem descriptions into executable mathematical optimization implementations using autonomous coding agents.

DynaWeb leverages model-based reinforcement learning to train web agents in simulated environments, enhancing the efficiency and scalability of autonomous internet navigation.

Develop a system that enhances LLM memory reliability in dynamic environments by detecting and realigning outdated or conflicting data.

A multi-agent framework for extracting procedural graphs from natural language with improved structural and logical accuracy.

Develop a system-theoretic framework for robust AI agent design to improve modularity and reliability.

Develop a method for learning observation and transition matrices in POMDPs beyond full-rank assumptions.