Selected work

Representative Papers

AgentR A Stateful and Recovery-Aware Software Architecture for LLM-based Auditable Workflows

Aug 15, 2026

This study addresses the critical challenges of state persistence and auditability in Large Language Model (LLM) applications by proposing AgentR, a stateful architecture. AgentR introduces a novel persistent state machine tailored for LLM workflows, integrating ACID-compliant cost logging and orphan task detection mechanisms. Through asynchronous orchestration, the system enables robust fault recovery and comprehensive end-to-end auditing. Experimental evaluations demonstrate that AgentR achieves a 99.2% task completion rate and a 4.3× parallel speedup. These results effectively validate the significant advantages of stateful design in enhancing the reliability, observability, and accountability of LLM-based systems.

0 citationsRead paper

PHISHREV: A Hybrid Machine Learning and Post-Hoc Non-monotonic Reasoning Framework for Context-Aware Phishing Website Classification

Apr 28, 2026

This work addresses the limitations of traditional phishing website detection methods, which rely on statistical machine learning models that lack contextual reasoning capabilities and are vulnerable to adversarial attacks. The authors propose a non-monotonic reasoning framework that integrates a machine learning classifier with Answer Set Programming (ASP), augmented by a belief revision layer that dynamically incorporates expert knowledge for context-aware decision refinement. This post-prediction module enables seamless integration of new domain knowledge without retraining the underlying model, operating with O(n) time complexity. Experimental results demonstrate that the reasoning component successfully corrects 5.08% of the classifier’s initial predictions, significantly enhancing the consistency and robustness of the detection outcomes.

0 citationsRead paper

CCAR: Intrinsic Robustness as an Emergent Geometric Property

Apr 18, 2026

Standard supervised learning often disregards the geometric structure of the feature space, leading to entangled representations that are highly sensitive to label noise and input perturbations. This work proposes Class-Conditional Activation Regularization (CCAR), which introduces a block-diagonal soft inductive bias to constrain the feature energy of each class within orthogonal subspaces, thereby explicitly constructing disentangled representations endowed with an intrinsic geometric skeleton. Theoretically, this study establishes, for the first time, a formal connection between geometric disentanglement and algorithmic stability, demonstrating that robustness arises from a well-structured feature space—achieved by maximizing the Fisher discriminant ratio. Extensive experiments show that CCAR significantly outperforms existing methods across multiple benchmarks involving label noise and input corruptions, confirming its effectiveness and enhanced robustness.

0 citationsRead paper

Affordable Precision Agriculture: A Deployment-Oriented Review of Low-Cost, Low-Power Edge AI and TinyML for Resource-Constrained Farming Systems

Mar 16, 2026

This study addresses the limited deployability of current precision agriculture systems in resource-constrained smallholder and underdeveloped settings due to their heavy reliance on cloud infrastructure. The work presents a systematic review of low-cost, low-power edge AI and TinyML applications tailored for real-world agricultural deployment, proposing a privacy-preserving hierarchical edge AI architecture that integrates microcontrollers (e.g., ESP32, STM32), single-board computers, and drone-assisted systems. It leverages techniques such as model quantization, pruning, and hardware-aware neural architecture search. The analysis reveals that approximately 50% of existing studies employ quantization for optimization, yet critical metrics like energy consumption and latency are inconsistently reported, hindering reproducibility. The feasibility of asymmetric training-inference architectures is validated, and key design principles are distilled to bridge the gap between research prototypes and deployable systems.

0 citationsRead paper

Synergizing a Decentralized Framework with LLM-Assisted Skill and Willingness-Aware Task Assignment for Volunteer Crowdsourcing

Mar 16, 2026

This study addresses the complex challenge of volunteer task assignment in crowdsourcing scenarios, where fine-grained skill heterogeneity, unstructured profile data, dynamically evolving willingness, and突发 task demands hinder effective allocation. To tackle these issues, this work proposes a novel framework that integrates large language model–based semantic preprocessing, an interpretable skill- and willingness-aware matching algorithm, and a blockchain-based execution mechanism. For the first time, the approach synergistically combines semantic intelligence, explainable matching, and decentralized execution to significantly enhance assignment quality and system trustworthiness without requiring on-chain optimization. Experimental results on multiple real-world resume datasets demonstrate a 42.3% improvement in task assignment utility and an increase in task coverage from 0.80 to 0.90.

0 citationsRead paper

Recent publications

Latest Papers

AgentR A Stateful and Recovery-Aware Software Architecture for LLM-based Auditable Workflows

Aug 15, 2026

This study addresses the critical challenges of state persistence and auditability in Large Language Model (LLM) applications by proposing AgentR, a stateful architecture. AgentR introduces a novel persistent state machine tailored for LLM workflows, integrating ACID-compliant cost logging and orphan task detection mechanisms. Through asynchronous orchestration, the system enables robust fault recovery and comprehensive end-to-end auditing. Experimental evaluations demonstrate that AgentR achieves a 99.2% task completion rate and a 4.3× parallel speedup. These results effectively validate the significant advantages of stateful design in enhancing the reliability, observability, and accountability of LLM-based systems.

0 citationsRead paper

PHISHREV: A Hybrid Machine Learning and Post-Hoc Non-monotonic Reasoning Framework for Context-Aware Phishing Website Classification

Apr 28, 2026

This work addresses the limitations of traditional phishing website detection methods, which rely on statistical machine learning models that lack contextual reasoning capabilities and are vulnerable to adversarial attacks. The authors propose a non-monotonic reasoning framework that integrates a machine learning classifier with Answer Set Programming (ASP), augmented by a belief revision layer that dynamically incorporates expert knowledge for context-aware decision refinement. This post-prediction module enables seamless integration of new domain knowledge without retraining the underlying model, operating with O(n) time complexity. Experimental results demonstrate that the reasoning component successfully corrects 5.08% of the classifier’s initial predictions, significantly enhancing the consistency and robustness of the detection outcomes.

0 citationsRead paper

CCAR: Intrinsic Robustness as an Emergent Geometric Property

Apr 18, 2026

Standard supervised learning often disregards the geometric structure of the feature space, leading to entangled representations that are highly sensitive to label noise and input perturbations. This work proposes Class-Conditional Activation Regularization (CCAR), which introduces a block-diagonal soft inductive bias to constrain the feature energy of each class within orthogonal subspaces, thereby explicitly constructing disentangled representations endowed with an intrinsic geometric skeleton. Theoretically, this study establishes, for the first time, a formal connection between geometric disentanglement and algorithmic stability, demonstrating that robustness arises from a well-structured feature space—achieved by maximizing the Fisher discriminant ratio. Extensive experiments show that CCAR significantly outperforms existing methods across multiple benchmarks involving label noise and input corruptions, confirming its effectiveness and enhanced robustness.

0 citationsRead paper

Affordable Precision Agriculture: A Deployment-Oriented Review of Low-Cost, Low-Power Edge AI and TinyML for Resource-Constrained Farming Systems

Mar 16, 2026

This study addresses the limited deployability of current precision agriculture systems in resource-constrained smallholder and underdeveloped settings due to their heavy reliance on cloud infrastructure. The work presents a systematic review of low-cost, low-power edge AI and TinyML applications tailored for real-world agricultural deployment, proposing a privacy-preserving hierarchical edge AI architecture that integrates microcontrollers (e.g., ESP32, STM32), single-board computers, and drone-assisted systems. It leverages techniques such as model quantization, pruning, and hardware-aware neural architecture search. The analysis reveals that approximately 50% of existing studies employ quantization for optimization, yet critical metrics like energy consumption and latency are inconsistently reported, hindering reproducibility. The feasibility of asymmetric training-inference architectures is validated, and key design principles are distilled to bridge the gap between research prototypes and deployable systems.

0 citationsRead paper

Synergizing a Decentralized Framework with LLM-Assisted Skill and Willingness-Aware Task Assignment for Volunteer Crowdsourcing

Mar 16, 2026

This study addresses the complex challenge of volunteer task assignment in crowdsourcing scenarios, where fine-grained skill heterogeneity, unstructured profile data, dynamically evolving willingness, and突发 task demands hinder effective allocation. To tackle these issues, this work proposes a novel framework that integrates large language model–based semantic preprocessing, an interpretable skill- and willingness-aware matching algorithm, and a blockchain-based execution mechanism. For the first time, the approach synergistically combines semantic intelligence, explainable matching, and decentralized execution to significantly enhance assignment quality and system trustworthiness without requiring on-chain optimization. Experimental results on multiple real-world resume datasets demonstrate a 42.3% improvement in task assignment utility and an increase in task coverage from 0.80 to 0.90.

0 citationsRead paper