ABSTRACT

This paper presents Sentient UI, a Flutter-based framework for emotion-aware user interface adaptation that operates entirely on-device. The framework introduces a modular architecture that separates emotion recognition, behavioral tracking, contextual awareness, adaptation logic, and UI integration. A proof-of-concept demonstration illustrates real-time visual adaptation under simulated emotional states, validating architectural feasibility while preserving user privacy and avoiding reliance on external cloud services. The work emphasizes engineering design and system architecture rather than large-scale user evaluation.

1. INTRODUCTION

Contemporary user interfaces remain largely static, providing limited responsiveness to variations in user emotion, interaction behavior, or situational context. While adaptive user interfaces have been explored extensively in academic research, their practical adoption is constrained by architectural complexity, privacy concerns, and integration overhead.

Sentient UI addresses these limitations by proposing a clean, modular architecture that enables emotion-aware adaptation directly within Flutter applications. By prioritizing on-device processing and explicit developer control, the framework aims to bridge the gap between adaptive UI research and real-world software engineering practice.

2. SYSTEM ARCHITECTURE

Sentient UI follows Clean Architecture principles and is structured into clearly defined layers, each with distinct responsibilities and interfaces. This separation of concerns promotes maintainability, extensibility, and independent evolution of system components.

• • Input Detection Layer: Captures emotional signals, behavioral interaction metrics, and contextual data using on-device mechanisms.
• • Signal Processing Layer: Performs normalization, temporal smoothing, and confidence-weighted fusion of multimodal signals.
• • Adaptation Logic Layer: Translates processed signals into concrete adaptation decisions using rule-based and multi-factor strategies.
• • UI Integration Layer: Applies adaptations through widget-level wrappers, preserving Flutter compatibility and declarative UI principles.

3. METHODOLOGY

Emotional input is modeled using lightweight on-device neural networks targeting Ekman's six basic emotions. Raw frame-level predictions are inherently noisy due to transient expressions and environmental variation. To address this instability, Sentient UI employs Bayesian sequential updating to aggregate predictions over time, producing temporally stable emotion confidence estimates.

Behavioral signals, including tap frequency and dwell time, as well as contextual signals such as time of day and device state, complement emotional input. These signals are normalized and fused using a confidence-weighted strategy to guide adaptation decisions while avoiding overreaction to transient events.

4. PROOF-OF-CONCEPT DEMONSTRATION

A Flutter-based proof-of-concept implementation demonstrates Sentient UI's ability to apply real-time visual adaptations under simulated emotional states. The demonstration focuses on validating architectural flow, adaptation triggering, and UI integration rather than production readiness.

Visual adaptations include color scheme transitions, contrast adjustments, and selective emphasis of interface elements, while maintaining consistent layout structure and functional behavior across emotional states.

5. KEY CONTRIBUTIONS

• • Design of a modular, Clean Architecture-based framework for adaptive UIs
• • On-device emotion stabilization using Bayesian sequential updating
• • Widget-level adaptation via decorator-based integration in Flutter
• • Privacy-preserving design with no external data transmission