LLM Agents and Applications

LLM agents for tool use, user simulation, education, medicine, finance, speech interaction, and micro-world environments.

Agents and applications
Tool agents User simulators AI patients Role-playing agents Market agents
LLM agents and applications project map

This project organizes papers where LLMs do more than answer: they plan with tools, simulate users, play roles, teach learners, act as standardized patients, participate in markets, or operate inside evolving environments.

Research Storyline

Plan
Use tools through multi-agent search

Smurfs studies how multiple agents can coordinate tool planning while keeping search context efficient.

Simulate
Generate interactive users and learners

UserGPT and PlatoLM turn LLMs into simulated users, giving dialogue systems richer multi-turn practice.

Practice
Deploy agents in education and medicine

EasyMED and AI standardized patients make clinical communication and reasoning practice repeatable and measurable.

World
Put agents into markets and micro-worlds

TwinMarket and MicroVerse move from isolated prompts to environments where agent actions change future states.

Paper Trail

Tools
Smurfs: Multi-Agent System using Context-Efficient DFSDT for Tool Planning

Uses multi-agent search for tool planning with better context efficiency.

Paper
User
Large Language Model as a User Simulator

Builds LLM user simulators for dialogue data generation, training, and evaluation.

Paper
Tutor
PlatoLM: Teaching LLMs via a Socratic Questioning User Simulator

Uses Socratic user simulation to improve multi-turn dialogue learning.

Paper
Patient
Human or LLM as Standardized Patients?

Compares AI standardized patients and human standardized patients for medical education.

Paper
Role
CRPO: Character-centric Group Relative Policy Optimization

Optimizes role-playing agents while preserving character fidelity and reasoning quality.

Paper
Market
TwinMarket

Simulates financial markets with LLM investor agents that carry beliefs, desires, intentions, and actions.

Project page
World
MicroVerse

Explores scientific micro-world simulation with hidden mechanisms and evolving states.

Project page

Application Clusters

Tool-use agents

Smurfs organizes tool planning as a multi-agent search problem rather than a single fragile ReAct trajectory.

Dialogue and education agents

User simulators, PlatoLM, and AI standardized patients use interaction to create practice, feedback, and evaluation loops.

Role-aware agents

CRPO studies how RL objectives should change when the agent must preserve persona, style, and character consistency.

World and market agents

TwinMarket, Economic World Models, and MicroVerse treat agents as actors inside environments that respond to their decisions.

Resource Map

Smurfs

Paper and ACL Anthology page for multi-agent tool planning.

Paper
EasyMED

AI standardized patient framework and medical education evaluation resources.

Repository
TwinMarket

Financial market simulation with LLM investor agents.

Project page
MicroVerse

Micro-world simulation for scientific environments.

Project page