Shimin Di
Publications
UrbanAgent: A Tool-Augmented Agent for Cross-System Urban Tasks
Modern cities rely on an increasing number of digital services to operate, but residents' daily needs are still difficult to meet. Services are fragmented and have little interoperability, placing a heavy operational burden on users. Existing digital platforms, urban foundation models, and intelligent assistants each address only isolated aspects of an urban task. But they struggle to reliably convert complex natural-language requests into executable cross-system workflows. We propose Urban-Agent, a tool-augmented agent framework for cross-system urban tasks. It couples the cognitive and reasoning capabilities of a large language model with a tool-set supporting code execution, API calls, and Model Context Protocol. Through one adaptive closed loop, it clarifies missing information before acting, grounds tool use in live observations, and aligns the final response with observed evidence and task constraints. To address the evaluation gap, we introduce Urban-Eval, a benchmark specifically designed for cross-system urban request. Unlike prior benchmarks that assess either general tool use or urban knowledge and reasoning, Urban-Eval evaluates both task results and execution quality, including required tool coverage, dependency validity, and evidence traceability. Experimental results indicate that Urban-Agent reaches a 71% task success rate, 10 points above the strongest baseline. This lead holds across GPT-5-mini, Gemini-2.5-flash, DeepSeek-V4-flash, and Qwen3-235B-A22B.
DREvo: Distilling Recalibrated Historical Experience for Harness Self-Evolution
Harness plays a critical role in large language model agent performance, and building a high-performing harness requires substantial expert effort. Therefore, recent research has increasingly explored harness self-evolution, which iteratively proposes, evaluates, and improves harnesses using historical trial experience. However, accumulated historical experience does not always translate into stable search guidance, and performance often fluctuates substantially across evolution iterations, making it difficult to reliably discover high-performing harnesses under a limited evolution budget. We identify two limitations in how existing harness self-evolution methods leverage historical experience: (1) Lack of dynamic reassessment of whether historical experience remains valid for the current harness, and (2) Lack of explicit mechanisms for translating valid historical experience into actionable search directions. To address these limitations, we propose a new harness self-evolution method, named DREvo, which integrates function-level evidence anchoring, state-dependent evidence recalibration, and role-conditioned search intent distillation to determine which historical evidence remains valid and where the harness should evolve next. Under limited evolution budgets, DREvo exhibits smoother evolution trajectories, achieves the highest accuracy on all five benchmarks, and delivers average gains of 16.2% and 14.2% over the evaluated baselines on domain reasoning and agentic tasks, respectively.
FactReview: Evidence-Grounded Reviews with Literature Positioning and Execution-Based Claim Verification
Peer review in machine learning is under growing pressure from rising submission volume and limited reviewer time. Most LLM-based reviewing systems read only the manuscript and generate comments from the paper's own narrative. This makes their outputs sensitive to presentation quality and leaves them weak when the evidence needed for review lies in related work or released code. We present FactReview, an evidence-grounded reviewing system that combines claim extraction, literature positioning, and execution-based claim verification. Given a submission, FactReview identifies major claims and reported results, retrieves nearby work to clarify the paper's technical position, and, when code is available, executes the released repository under bounded budgets to test central empirical claims. It then produces a concise review and an evidence report that assigns each major claim one of five labels: Supported, Supported by the paper, Partially supported, In conflict, or Inconclusive. In a case study on CompGCN, FactReview reproduces results that closely match those reported for link prediction and node classification, yet also shows that the paper's broader performance claim across tasks is not fully sustained: on MUTAG graph classification, the reproduced result is 88.4%, whereas the strongest baseline reported in the paper remains 92.6%. The claim is therefore only partially supported. More broadly, this case suggests that AI is most useful in peer review not as a final decision-maker, but as a tool for gathering evidence and helping reviewers produce more evidence-grounded assessments. The code is public at https://github.com/DEFENSE-SEU/Review-Assistant.
Learning to Compose for Cross-domain Agentic Workflow Generation
Automatically generating agentic workflows -- executable operator graphs or codes that orchestrate reasoning, verification, and repair -- has become a practical way to solve complex tasks beyond what single-pass LLM generation can reliably handle. Yet what constitutes a good workflow depends heavily on the task distribution and the available operators. Under domain shift, current systems typically rely on iterative workflow refinement to discover a feasible workflow from a large workflow space, incurring high iteration costs and yielding unstable, domain-specific behavior. In response, we internalize a decompose-recompose-decide mechanism into an open-source LLM for cross-domain workflow generation. To decompose, we learn a compact set of reusable workflow capabilities across diverse domains. To recompose, we map each input task to a sparse composition over these bases to generate a task-specific workflow in a single pass. To decide, we attribute the success or failure of workflow generation to counterfactual contributions from learned capabilities, thereby capturing which capabilities actually drive success by their marginal effects. Across stringent multi-domain, cross-domain, and unseen-domain evaluations, our 1-pass generator surpasses SOTA refinement baselines that consume 20 iterations, while substantially reducing generation latency and cost.
PeroMAS: A Multi-agent System of Perovskite Material Discovery
As a pioneer of the third-generation photovoltaic revolution, Perovskite Solar Cells (PSCs) are renowned for their superior optoelectronic performance and cost potential. The development process of PSCs is precise and complex, involving a series of closed-loop workflows such as literature retrieval, data integration, experimental design, and synthesis. However, existing AI perovskite approaches focus predominantly on discrete models, including material design, process optimization,and property prediction. These models fail to propagate physical constraints across the workflow, hindering end-to-end optimization. In this paper, we propose a multi-agent system for perovskite material discovery, named PeroMAS. We first encapsulated a series of perovskite-specific tools into Model Context Protocols (MCPs). By planning and invoking these tools, PeroMAS can design perovskite materials under multi-objective constraints, covering the entire process from literature retrieval and data extraction to property prediction and mechanism analysis. Furthermore, we construct an evaluation benchmark by perovskite human experts to assess this multi-agent system. Results demonstrate that, compared to single Large Language Model (LLM) or traditional search strategies, our system significantly enhances discovery efficiency. It successfully identified candidate materials satisfying multi-objective constraints. Notably, we verify PeroMAS's effectiveness in the physical world through real synthesis experiments.
Sci-VLA: Agentic VLA Inference Plugin for Long-Horizon Tasks in Scientific Experiments
Robotic laboratories play a critical role in autonomous scientific discovery by enabling scalable, continuous experimental execution. Recent vision-language-action (VLA) models offer a promising foundation for robotic laboratories. However, scientific experiments typically involve long-horizon tasks composed of multiple atomic tasks, posing a fundamental challenge to existing VLA models. While VLA models fine-tuned for scientific tasks can reliably execute atomic experimental actions seen during training, they often fail to perform composite tasks formed by reordering and composing these known atomic actions. This limitation arises from a distributional mismatch between training-time atomic tasks and inference-time composite tasks, which prevents VLA models from executing necessary transitional operations between atomic tasks. To address this challenge, we propose an Agentic VLA Inference Plugin for Long-Horizon Tasks in Scientific Experiments. It introduces an LLM-based agentic inference mechanism that intervenes when executing sequential manipulation tasks. By performing explicit transition inference and generating transitional robotic action code, the proposed plugin guides VLA models through missing transitional steps, enabling reliable execution of composite scientific workflows without any additional training. This inference-only intervention makes our method computationally efficient, data-efficient, and well-suited for open-ended and long-horizon robotic laboratory tasks. We build 3D assets of scientific instruments and common scientific operating scenes within an existing simulation environment. In these scenes, we have verified that our method increases the average success rate per atomic task by 42\% during inference. Furthermore, we show that our method can be easily transferred from the simulation to real scientific laboratories.
Guided by Trajectories: Repairing and Rewarding Tool-Use Trajectories for Tool-Integrated Reasoning
Tool-Integrated Reasoning (TIR) enables large language models (LLMs) to solve complex tasks by interacting with external tools, yet existing approaches depend on high-quality synthesized trajectories selected by scoring functions and sparse outcome-based rewards, providing limited and biased supervision for learning TIR. To address these challenges, in this paper, we propose AutoTraj, a two-stage framework that automatically learns TIR by repairing and rewarding tool-use trajectories. Specifically, in the supervised fine-tuning (SFT) stage, AutoTraj generates multiple candidate tool-use trajectories for each query and evaluates them along multiple dimensions. High-quality trajectories are directly retained, while low-quality ones are repaired using a LLM (i.e., LLM-as-Repairer). The resulting repaired and high-quality trajectories form a synthetic SFT dataset, while each repaired trajectory paired with its original low-quality counterpart constitutes a dataset for trajectory preference modeling. In the reinforcement learning (RL) stage, based on the preference dataset, we train a trajectory-level reward model to assess the quality of reasoning paths and combine it with outcome and format rewards, thereby explicitly guiding the optimization toward reliable TIR behaviors. Experiments on real-world benchmarks demonstrate the effectiveness of AutoTraj in TIR.
ACR: Adaptive Context Refactoring via Context Refactoring Operators for Multi-Turn Dialogue
Large Language Models (LLMs) have shown remarkable performance in multi-turn dialogue. However, in multi-turn dialogue, models still struggle to stay aligned with what has been established earlier, follow dependencies across many turns, and avoid drifting into incorrect facts as the interaction grows longer. Existing approaches primarily focus on extending the context window, introducing external memory, or applying context compression, yet these methods still face limitations such as \textbf{contextual inertia} and \textbf{state drift}. To address these challenges, we propose the \textbf{A}daptive \textbf{C}ontext \textbf{R}efactoring \textbf{(ACR)} Framework, which dynamically monitors and reshapes the interaction history to mitigate contextual inertia and state drift actively. ACR is built on a library of context refactoring operators and a teacher-guided self-evolving training paradigm that learns when to intervene and how to refactor, thereby decoupling context management from the reasoning process. Extensive experiments on multi-turn dialogue demonstrate that our method significantly outperforms existing baselines while reducing token consumption.