Applications in the 3D Computing Continuum, which unifies edge, cloud, and space, require combining multiple AI tasks such as object detection, time-series analytics, and natural language processing into Compound AI systems. These systems must satisfy stringent Service Level Objectives (SLOs) on accuracy, latency, and cost. A key mechanism for maintaining SLO compliance of Compound AI systems is runtime model selection, where AI models are dynamically switched for each workflow task. However, existing distributed and compound AI frameworks do not natively support runtime model selection. We present PLAIground, a framework that enables runtime model selection for Compound AI systems. PLAIground introduces Compoundable AI Model (CAIM) abstraction, which decouples task semantics from AI model implementations via Task and Data Contracts, enabling model switching without workflow changes. Additionally, PLAIground introduces Pixie, an SLO-driven runtime model selection algorithm, which dynamically selects the most suitable model for each task during execution. Our evaluation on two realistic Compound AI workflows demonstrates that Pixie achieves up to 91.3
Stuttering can make everyday conversations challenging, especially in situations that involve unfamiliar people or social pressure. While traditional therapy provides structured support, many individuals struggle to apply those techniques in real-world scenarios. Digital speech tools exist, but most focus on repetitive drills and rarely provide realistic, contextdriven speaking practice. We present EaseTalk, an AI-driven mobile application that allows users to practice real-life conversational scenarios in a supportive, self-paced environment. The app uses speech recognition and prompt-engineered large language models (LLMs) to detect speech disfluencies such as repetitions, prolongations and blocks. EaseTalk aims to empower users through independent, scenario-based speech practice with potential applications extending beyond stuttering into areas like social anxiety, interview preparation and language learning.
The explosive growth of the Internet of Things (IoT) demands security mechanisms that adapt to emerging threats. Telemetry meets this need by fusing federated learning, explainable AI, and privacy-preserving analytics into a multi-layer monitoring framework. Lightweight agents such as r-Monitoring impose only 0.27 % CPU overhead at device level, while the BACON federated anomaly detector safeguards system traffic. On a 21-sensor industrial robot, TELEMETRY's Nokia pipeline flagged subtle speed anomalies with 73% accuracy; within the NF-ToN-IoT corpus, BACON differentiated benign from malicious flows with 96% accuracy and markedly fewer false positives than signature baselines. The Misuse Detection Toolkit ensemble further achieved 97.7% validation accuracy across 50 training epochs, underscoring the framework's adaptability. Together, these layers cut detection latency and reduce on-device resource use, illustrating how ML-driven, federated monitoring can harden next-generation IoT deployments. Ongoing work explores graph-based analytics, adaptive models, and more scalable federation schemes.
Autonomous vehicles (AVs) are becoming integral to on-demand micro transit, offering the potential for safer, efficient, and sustainable transportation. However, AV deployment faces several challenges, including the lack of suitable roadways, varying travel conditions. Traditional routers prioritize speed and not reliability, leading to unpredictable operations and complications in planning. To address these, we introduce AVATAR, an autonomy-aware routing framework that prioritizes dependable, low-variance routes. Our approach encodes multiple objectives including road speed, speed variability, zoning areas, pedestrian encounters, and operator preferred roadways into edge-level routing engines. Objective optimized routes are generated, then scored using a multi-criteria decision-making process. User-configurable preference profiles, allow operators to define a balance between reliability and speed. AVATAR is a data-driven framework that supports both real-time AV operations and offline analysis, enabling transit operators to assess and refine routing strategies. Our experiments using real-world data from Silicon Valley, California, and Yokohama, Japan show that our approach significantly improves AV reliability and performance and advances the sustainable and scalable integration of AVs into future transportation networks.
With the growing adoption of location-based services, ensuring user privacy without compromising data utility has become a critical research focus. In this paper, we investigate the impact of different statistical noise distributions-Laplace, Gaussian, Exponential, Cauchy, and Uniform, on trajectory anonymization across different transportation modes using Differential Privacy. We propose a modality-wise anonymization approach and assess its effectiveness through two down-stream tasks utility metric: (i) counting the number of unique users within a predefined area, and (ii) predicting modes of transport. Experimental results on the two public datasets Geolife and Roma taxi reveal that our modality based anonymization retains high utility, achieving up to 5% improvement in micro-F1 scores compared to single-noise based anonymization and preserves user counts (deviates by 4 users) closely as of original data. These findings emphasize that downstream tasks play a pivotal role in designing privacy-preserving mechanisms while maintaining the practical usability of mobility data.
