Which on-device LLM best fits a given user? 2B-class edge models exhibit user-specific behavioral variation: one may better align with genre-sensitive recommendations, another with procedural instructions, and no single model dominates across all users or domains. We present PEARL (Profile-based Explainable Agent for edge-model Recommendation via Latent Decomposition), which selects the best-fitting 2B-class edge LLM from a structured user profile and generates a natural-language explanation grounded in interpretable latent alignment dimensions. Its core, Explainable Latent Decomposition (ELD), maps per-model profile-alignment fingerprints into a shared low-correlation latent subspace for direct profile-to-model matching. On PersonaLens (200 profiles, 8,521/1,974 TSD/TMD dialogues), PEARL achieves P = 2.23 \pm .02 on TSD, outperforming the best single-fixed edge model by +0.11 and closing 35.0% of the Oracle-UB gap; on TMD, PEARL achieves P = 2.14 \pm .03 (+0.11, 34.0% Oracle-UB gap closure). A perturbation analysis shows that masking the top-1 ELD dimension alters 60% of recommendations (Δ P = -0.43), with a modest 1.3× specificity ratio over a random-dimension control.

Modern web interfaces increasingly support complex decision workflows, such as travel planning and multi-criteria selection, yet remain largely static and insensitive to users' moment-to-moment cognitive states during interaction. Travel planning, in particular, requires users to synthesize dispersed information under multiple constraints, making it a representative high-load interactive decision task. This study presents MACA (Multi-Agent Cognitive Adaptation), a framework that enables real-time cognitive adaptation in web-based decision environments by integrating hierarchical Monte Carlo Tree Search with a Planner–Critic–Executor multi-agent architecture. MACA continuously estimates users' emotional and attentional states using facial expression analysis (ResEmoteNet) and gaze stability tracking (MediaPipe), and uses these signals to regulate agent collaboration, reasoning depth, and feedback pacing during interaction. We evaluated MACA in a 2×2 within-subject study (N = 30) comparing Single versus Multi-agent and Fixed versus Adaptive configurations. Results show that the Multi-Adaptive condition significantly improved decision quality (F(3,116) = 2.96, p = 0.035) while reducing mental effort (F(3,116) = 2.82, p = 0.042), yielding a 10.7% gain in decision efficiency without increasing cognitive burden. These findings demonstrate that multimodal user-state sensing combined with cooperative multi-agent reasoning can enhance interactive web-based decision making while maintaining user well-being.

This study introduces an informational and empathetic chatbot for childhood cancer survivors. As the survival rates for childhood cancer around the world have increased, survivors often face various psychosocial challenges during and after cancer treatment. However, they rarely seek support from psychosocial professionals due to the low availability of resources and stigma toward cancer survivors in countries like South Korea. This study aimed to develop a chatbot tailed to the unique characteristics of childhood cancer survivors in need of informational and emotional support. Given the limited availability of empirical data on childhood cancer survivors, quotes from survivors were gathered from academic articles and social media, then large language models were employed to generate appropriate responses. Furthermore, we incorporated domain learning techniques to ensure a more tailored and suitable model for addressing the needs of survivors.