Selected work

Representative Papers

AIvailable: A Software-Defined Architecture for LLM-as-a-Service on Heterogeneous and Legacy GPUs

Nov 06, 2025

Deploying high-performance large language model (LLM) inference services remains challenging in resource-constrained environments—such as academic labs and SMEs—especially when leveraging heterogeneous, legacy GPU hardware (NVIDIA/AMD). Method: This paper proposes a low-overhead LLM-as-a-Service platform designed for heterogeneous legacy GPUs. It employs a software-defined AI architecture featuring a novel VRAM-aware dynamic model allocation and rescheduling mechanism, abstracting hardware heterogeneity to enable fully GPU-accelerated, CPU-fallback-free cross-vendor inference. The platform integrates secure request routing, lightweight load balancing, and distributed monitoring. Contribution/Results: To our knowledge, this is the first system enabling highly available, elastic inference of open-source LLMs on mixed legacy GPU clusters, significantly improving VRAM utilization. It offers a unified API supporting multiple models, lowering deployment barriers. Experiments show a 2.3× increase in VRAM utilization over baseline approaches, 99.8% service availability, and substantial reduction in LLMaaS operational cost.

0 citationsRead paper

HaLert: A Resilient Smart City Architecture for Post-Disaster Based on Wi-Fi HaLow Mesh and SDN

Jul 10, 2025

To address disrupted emergency communication caused by traditional infrastructure failure after disasters, this paper proposes a resilient emergency communication architecture leveraging existing smart-city IoT infrastructure. Methodologically, it innovatively integrates Wi-Fi HaLow (IEEE 802.11s) multi-hop mesh networking, Software-Defined Networking (SDN), and a LoRa-based flooding control plane: the former enables high-reliability, low-latency multimodal interaction (audio/video, text, location), while the latter supports wide-area, low-power, adaptive remote network management. Experimental evaluations in complex urban environments demonstrate that the Wi-Fi HaLow mesh achieves average end-to-end latency of 15–54.8 ms and throughput in the hundreds of kbps; LoRa control-message delivery success rate reaches 94.96%. The architecture significantly enhances post-disaster communication robustness, manageability, and heterogeneous system interoperability.

0 citationsRead paper

Project Riley: Multimodal Multi-Agent LLM Collaboration with Emotional Reasoning and Voting

May 26, 2025

This work addresses the limitations in multimodal perception, emotion modeling, and response coherence in affective conversational systems. We propose a personified multi-agent architecture inspired by *Inside Out*, comprising five emotion-specific agents—Joy, Sadness, Fear, Anger, and Disgust. The framework integrates multimodal large language models (text + vision), multi-turn negotiation, self-reflective iterative optimization, majority-voting fusion, and RAG-enhanced retrieval to enable emotion-driven dynamic perspective integration and logically coherent response generation. Key contributions include: (1) the first emotion-role-based multi-agent collaborative reasoning framework; (2) support for offline, lightweight deployment; and (3) an emergency-specialized variant, Armando, featuring cumulative context tracking and emotion-calibrated factual retrieval. User evaluations demonstrate statistically significant improvements over baselines in emotional appropriateness, expressive clarity, and naturalness, while maintaining real-time responsiveness.

0 citationsRead paper

HashKitty: Distributed Password Analysis

May 09, 2025

Addressing challenges in practical cryptanalysis—including poor adaptability to heterogeneous environments, inefficient task scheduling, and suboptimal resource utilization—this paper proposes and implements a lightweight distributed password cracking platform built upon hashcat. The platform introduces a novel WebSocket-driven modular architecture, enabling dynamic onboarding and unified management of heterogeneous nodes across operating systems, CPU/GPU architectures, and vendor-specific hardware. Leveraging coordinated Web APIs and microservices, it achieves intelligent hash task distribution, real-time status synchronization, and adaptive load balancing. Experimental evaluation demonstrates that the platform significantly improves large-scale hash cracking throughput (3.2×–5.8× higher than single-node execution) and resource utilization in mixed-hardware environments. It thus delivers a highly compatible, scalable, and production-ready distributed solution for operational cryptanalysis.

0 citationsRead paper

Recent publications

Latest Papers

AIvailable: A Software-Defined Architecture for LLM-as-a-Service on Heterogeneous and Legacy GPUs

Nov 06, 2025

Deploying high-performance large language model (LLM) inference services remains challenging in resource-constrained environments—such as academic labs and SMEs—especially when leveraging heterogeneous, legacy GPU hardware (NVIDIA/AMD). Method: This paper proposes a low-overhead LLM-as-a-Service platform designed for heterogeneous legacy GPUs. It employs a software-defined AI architecture featuring a novel VRAM-aware dynamic model allocation and rescheduling mechanism, abstracting hardware heterogeneity to enable fully GPU-accelerated, CPU-fallback-free cross-vendor inference. The platform integrates secure request routing, lightweight load balancing, and distributed monitoring. Contribution/Results: To our knowledge, this is the first system enabling highly available, elastic inference of open-source LLMs on mixed legacy GPU clusters, significantly improving VRAM utilization. It offers a unified API supporting multiple models, lowering deployment barriers. Experiments show a 2.3× increase in VRAM utilization over baseline approaches, 99.8% service availability, and substantial reduction in LLMaaS operational cost.

0 citationsRead paper

HaLert: A Resilient Smart City Architecture for Post-Disaster Based on Wi-Fi HaLow Mesh and SDN

Jul 10, 2025

To address disrupted emergency communication caused by traditional infrastructure failure after disasters, this paper proposes a resilient emergency communication architecture leveraging existing smart-city IoT infrastructure. Methodologically, it innovatively integrates Wi-Fi HaLow (IEEE 802.11s) multi-hop mesh networking, Software-Defined Networking (SDN), and a LoRa-based flooding control plane: the former enables high-reliability, low-latency multimodal interaction (audio/video, text, location), while the latter supports wide-area, low-power, adaptive remote network management. Experimental evaluations in complex urban environments demonstrate that the Wi-Fi HaLow mesh achieves average end-to-end latency of 15–54.8 ms and throughput in the hundreds of kbps; LoRa control-message delivery success rate reaches 94.96%. The architecture significantly enhances post-disaster communication robustness, manageability, and heterogeneous system interoperability.

0 citationsRead paper

Project Riley: Multimodal Multi-Agent LLM Collaboration with Emotional Reasoning and Voting

May 26, 2025

This work addresses the limitations in multimodal perception, emotion modeling, and response coherence in affective conversational systems. We propose a personified multi-agent architecture inspired by *Inside Out*, comprising five emotion-specific agents—Joy, Sadness, Fear, Anger, and Disgust. The framework integrates multimodal large language models (text + vision), multi-turn negotiation, self-reflective iterative optimization, majority-voting fusion, and RAG-enhanced retrieval to enable emotion-driven dynamic perspective integration and logically coherent response generation. Key contributions include: (1) the first emotion-role-based multi-agent collaborative reasoning framework; (2) support for offline, lightweight deployment; and (3) an emergency-specialized variant, Armando, featuring cumulative context tracking and emotion-calibrated factual retrieval. User evaluations demonstrate statistically significant improvements over baselines in emotional appropriateness, expressive clarity, and naturalness, while maintaining real-time responsiveness.

0 citationsRead paper

HashKitty: Distributed Password Analysis

May 09, 2025

Addressing challenges in practical cryptanalysis—including poor adaptability to heterogeneous environments, inefficient task scheduling, and suboptimal resource utilization—this paper proposes and implements a lightweight distributed password cracking platform built upon hashcat. The platform introduces a novel WebSocket-driven modular architecture, enabling dynamic onboarding and unified management of heterogeneous nodes across operating systems, CPU/GPU architectures, and vendor-specific hardware. Leveraging coordinated Web APIs and microservices, it achieves intelligent hash task distribution, real-time status synchronization, and adaptive load balancing. Experimental evaluation demonstrates that the platform significantly improves large-scale hash cracking throughput (3.2×–5.8× higher than single-node execution) and resource utilization in mixed-hardware environments. It thus delivers a highly compatible, scalable, and production-ready distributed solution for operational cryptanalysis.

0 citationsRead paper