AI Sens., Volume 2, Issue 2 (June 2026) – 4 articles

The increasing expansion of ubiquitous sensing systems has created large streams of time-series data that are difficult for non-technical users to interpret. Large Language Models (LLMs) offer a promising interface for transforming sensor data into natural language insights, particularly in distributed environments where users may lack familiarity with data analysis. However, models optimized for text generation often struggle to interpret raw time-series signals, producing responses that are generic, inaccurate, or poorly grounded in the data. This study evaluates a prompt structure based on the Retrieval-Augmented Generation (RAG) framework for interpreting sensor-derived time-series data from water-consumption monitoring systems installed in household storage tanks. The prompt integrates statistical summaries, sensor metadata, and contextual information about household water-use practices. Performance is evaluated using synthetic datasets representing a year of tank water-consumption measurements and a rubric-based evaluation framework applied by three independent language-model evaluators. Results show that augmenting prompts with structured contextual information improves the clarity and grounding of language model responses to sensor time-series data, increasing evaluation scores and reducing failure modes such as hallucination, contradiction with the data, and misuse of contextual information, as assessed by independent evaluator models. These findings highlight the potential of structured contextual prompting to support locally deployed language models that produce reliable and actionable interpretations of sensor time-series data. Full article

Clinical monitoring of chronic liver disease (CLD) is currently hindered by the invasiveness of conventional biopsies. While breath-borne volatile organic compound (VOC) analysis offers a promising non-invasive alternative, the metabolic profiles of sedentary populations often overlap significantly with those of healthy individuals, making latent pathologies difficult to identify. To overcome this high-resolution diagnostic challenge, this study developed an integrated framework that couples high-performance semiconductor sensing technology with a machine learning-based analytical baseline. During the biomarker screening phase, GC-MS was utilized to analyze over 2000 VOCs, identifying 20 markers associated with CLD. These were further optimized into a robust feature panel including ammonia, isoprene, dimethyl sulfide (DMS), and limonene. For several critical metabolic features exhibiting high diagnostic potential, preliminary identifications were conducted by referencing NIST database matches and relevant literature. To maintain analytical rigor and account for the inherent complexity of trace volatile metabolites in biological samples, these signals are treated as putative metabolic features and characterized by their retention times. Regarding hardware, an InN-based sensor with Pt-AlN surface modification was fabricated, achieving a limit of detection (LOD) for ammonia below 0.2 ppm. Crucially, while the InN sensor was validated for specific core markers such as ammonia, the current AI classification model is trained on a refined 7-VOC panel derived from the comprehensive GC-MS data. To resolve diagnostic overlaps, a three-state dynamic sampling protocol (resting, exercise, and recovery) was implemented to isolate biomarkers that remain physiologically stable. By integrating multi-dimensional VOC features (e.g., isoprene and DMS) with sensor-validated data through DBSCAN and Random Forest algorithms, the framework successfully captured non-linear metabolic fingerprints. Machine learning results confirm that the framework effectively distinguished sedentary controls from CLD patients, achieving a macro-average AUC of 0.96. This integration provides a high-precision technical pathway for early-stage liver disease screening. Full article

For electromagnetic (EM)-driven design, we summarize surrogate modeling and inverse design based on full-wave simulations and, when available, measurements, use adjacent-domain EM exemplars only as methodological templates where direct closed-loop RF design-to-recognition evidence remains limited, and explain why reported speedups depend on the sampled design domain, sample density near feasibility boundaries, and how constraint-boundary checks and final verification are implemented. For RF sensing and recognition, we discuss how the learned “signature” is shaped by the measurement chain (hardware, placement, synchronization, calibration, and preprocessing). We then outline an author-synthesized set of checkable evaluation and reporting items, including explicit domains/constraints, complete sensing metadata, cross-condition tests, and edge deployment evidence, where deployability is defined at the full-pipeline level and requires measured latency, throughput, peak memory, energy per inference, preprocessing overhead, runtime, numerical precision, and an explicit statement of whether communication or offloading is included. Full article