Resource allocation in software projects is a critical challenge, with inefficiencies in handling dynamic constraints such as skill mismatches, evolving task dependencies, and budget overruns leading to widespread cost and schedule overruns. To address these issues, this paper introduces CAMP-GNN, a novel constraint-aware graph neural network (GNN) framework designed to optimize resource allocation in software projects, addressing challenges like skill mismatches, evolving task dependencies, and budget overruns. These issues often lead to significant cost and schedule overruns, and CAMP-GNN aims to resolve them by dynamically adapting to project constraints. The framework integrates urgency, skill alignment, and budget slack directly into its message-passing layers, allowing for more precise and context-aware task-resource assignments. A key contribution is SoftStat, the first structured dataset that combines resource pools, allocations, and project details, enabling the modeling of complex task-resource dependencies and financial/temporal constraints. CAMP-GNN also introduces domain-specific message aggregation rules to unify urgency, skill alignment, and budget slack, advancing GNNs beyond basic relational modeling. A novel preprocessing pipeline is proposed to convert raw project timelines, skill dependencies, and budget thresholds into machine-learnable features, ensuring effective integration with computational models. Validated on both synthetic and real-world datasets, CAMP-GNN demonstrates superior performance with a 0.12 MAE, RMSE of 0.18 in allocation, F1 of 0.89 in budget violations, and 085 in deadline violations, 18% cost savings, and a 27% reduction in deadline breaches, outperforming existing methods. Future work will focus on scaling the framework for large-scale projects and improving interpretability for broader adoption.

Emotion recognition from speech is crucial for advancing human-computer interactions, enabling more natural and empathetic communication. This study proposes a novel Speech Emotion Recognition (SER) framework that integrates Convolutional Neural Networks (CNNs) and transformer-based architectures to capture local and contextual speech features. The model demonstrates strong classification performance, particularly for prominent emotions such as anger, sadness, and happiness. However, challenges persist in detecting less frequent emotions, such as surprise and calm, highlighting areas for improvement. The limitations of current datasets, such as limited linguistic diversity, are discussed. The findings underscore the model's robustness and identify avenues for future enhancement, such as incorporating more diverse datasets and employing techniques such as transfer learning. Future work will explore multimodal approaches and real-time implementation on edge devices to improve the system's adaptability in real-world scenarios.

Accreditation and quality assurance in technical and vocational education and training (TVET) remain largely dependent on manual review and expert judgment. This study presents AccreditNet, an explainable tabular AI model embedded in a broader accreditation decision support framework. The experimentally evaluated component combines per-feature tokenization, multi-head self-attention, hierarchical criterion fusion, class-balanced focal loss, integrated gradients, and descriptive performance gap scoring. Evaluation is conducted against a structured NAAC benchmark that includes criterion scores and institutional attributes. Because the official NAAC grade is derived from criterion-level scoring, the benchmark is interpreted as a test of how well the model represents the criterion-to-grade mapping rather than as evidence of an independent predictive relationship. On the reported held-out split, AccreditNet achieved 93.72% accuracy, 92.84% macro F1-score, and 0.9706 AUC-ROC. OCR-based ingestion, NLP/LLM-assisted evidence mapping, automated reporting, and Saudi deployment are proposed operational extensions and were not evaluated in the present experiments. The results support further study of explainable accreditation analytics and decision support, subject to validation on local data.

MSG-VA is a hierarchical graph pipeline with analyst-in-the-loop visual analytics for real-time network threat detection. Under a defined threat model, adversaries manipulating host–to–host communication patterns—assumptions include flow-level telemetry and a sub-second latency budget. Entities are aggregated by subnet/role and time to form multi-scale graphs. A lightweight GNN computes embeddings and flags anomalies using adaptive thresholding. Alerts and provenance paths are surfaced in an interactive dashboard. On CICIDS2017 and UNSW-NB15, MSG-VA attains F1 94.8%, Accuracy 95.0%, and FPR 5.6%, outperforming tuned RF and DNN baselines under identical, leakage-free splits. An ablation study isolating the multi-scale design shows a 2–3× reduction in compute, with ≤ 0.5 pp F1 loss, and a 10-task analyst study indicates a 34% faster triage. A reproducible end-to-end benchmark reports a median alert latency of 95.7 ms (including graph construction, inference, and visualization push) and a model-inference throughput of ~ 250,000 nodes/s. Code, exact splits, and configurations are released to enable replication. Contributions: (i) a concrete multi-scale coarsening policy tied to network roles and time; (ii) a compact GNN with adaptive thresholding that improves detection at low FPR; (iii) a real-time visual analytics workflow validated by ablations, zero-day/cross-dataset tests, and a small user study.

High-Frequency Trading (HFT) dashboards require rapid reception, aggregation, and visualization of order book and trade update streams that may arrive at multi-million message rates. Conventional CPU-based and CPU-GPU hybrid visualization pipelines can suffer from significant delays during periods of burst due to CPU-mediated rendering, synchronization, kernel launch overhead, and copies on the host. This paper presents a visualization pipeline that is entirely resident on the graphics processor with zero-copy access to NIC accessible pinned buffers, persistent CUDA processing, fused stage execution of the parse-aggregate pipeline, and persistent CUDA OpenGL buffer interoperation. The goal is not to reach production status but rather to see whether host-to-host data movement can be decreased and whether the stages of GPU processing can be consolidated to improve latency, throughput and frame cadence in controlled HFT-style workloads. The evaluated workstation achieved a mean ingest-to-pixel latency of 6.3 ms using the proposed design compared to 29.4 ms for the current design, with sustained throughput of 10.2 million messages per second, which is 20 times greater than the current design, and a steady-state range of 185 to 192 frames per second with a burst floor of 178 frames per second for the proposed design. The improvement observed can be attributed to both the zero-copy ingestion and fused persistent kernel execution. Based on the obtained results, the proposed method of use of this technique in the implementation of real-time financial visualization under the proposed conditions is possible. More general testing is still required on other NICs, other generations of GPUs and PCIe configurations, workload traces, and actual exchange feeds.