AI systems, including large language models (LLMs) and agentic applications, must be accurate, reliable, and scalable while operating under constraints on latency, memory, energy, bandwidth, and privacy. Our research develops resource-aware methods for efficient training, fine-tuning, and inference across cloud, edge, and device environments. We study compression, quantization, distillation, pruning, caching, adaptive inference, and parameter-efficient fine-tuning to reduce computational cost. We also design distributed, decentralized, and federated AI solutions that support interactive, privacy-aware, and context-sensitive intelligent services close to users and data sources.

Modern search systems must provide accurate, low-latency results across massive, heterogeneous collections, increasingly using neural models, semantic search, and retrieval-augmented generation (RAG). Our research focuses on scalable retrieval based on learned representations, with a focus on efficient indexing, compression, storage, and querying of dense and sparse embeddings. We combine vector search, approximate nearest-neighbor methods, in-memory data management, and adaptive query processing to design ranking, re-ranking, hybrid retrieval, and RAG pipelines that balance effectiveness, efficiency and scalability.

Modern digital services operate across cloud, edge, device, and federated environments. Our research focuses on building scalable and adaptive platforms that bring computation and data closer to users, reducing latency, bandwidth use, energy consumption, and cost. To achieve this, we develop decentralized and AI-enhanced orchestration methods for dynamic service placement, scheduling, resource allocation, and migration across heterogeneous infrastructures. By adapting to workload changes, resource conditions, mobility, and demand, our solutions improve the performance and efficiency of latency-sensitive, data-intensive, and AI-enabled applications.

Our research develops AI-driven methods for analyzing mobility, social, textual, and spatio-temporal data. We combine semantic technologies, machine learning, data mining, large language models, and knowledge-based approaches to model human behavior, movement patterns, urban dynamics, and online interactions. We study semantic trajectory analysis, context-aware services, large-scale mobility analytics, semantic enrichment, and LLM-based understanding of text and social media. These methods support intelligent applications in smart mobility, transportation, tourism, urban computing, and knowledge-based content understanding.