St. Thomas' College of Engineering and Technology

Conversational agents designed for emotionally supportive interactions face challenges in balancing affective responsiveness, computational efficiency, and safety in communication. Prior approaches frequently depend on large-scale models, handcrafted affective objectives, or reinforcement learning from human feedback, which can limit scalability and interpretability. This work presents a lightweight, domain-adapted dialogue generation system based on the T5-small architecture, fine-tuned on MentalChat16K, a curated corpus of real and synthetic emotional-support conversations. The proposed model operates without reinforcement learning or emotion-specific training objectives, yet demonstrates encouraging alignment with affective cues and fluent response generation within the evaluated dataset. Empirical evaluation shows improvements over zero-shot and fine-tuned GPT-2 baselines, achieving BLEU (32.14), ROUGE-L (44.72), and BERTScore-F1 (85.11). Expert human assessments indicated high ratings in coherence, emotional appropriateness, and contextual relevance, with substantial inter-rater agreement. Qualitative error analysis indicated generally conservative and context-aware responses within the evaluated sample. During manual review of this sample, no factual hallucinations, medical overreach, or overtly unsafe responses were observed; however, systematic safety benchmarking was beyond the scope of the present study. This study provides initial evidence that compact transformer-based models, when adapted to domain-specific corpora and evaluated under controlled conditions, can support efficient and affectively appropriate dialogue generation in emotionally supportive non-clinical settings, while requiring further safety validation before broader real-world deployment.

Fibrosis involves the development and hardening of fibrous connective tissue due to organ damage and marks the initial stage of cancer. Detecting fibrosis early is crucial to prevent further cell damage. This paper introduces a method for identifying fibrosis-affected regions in zoomed, colorful images of rat liver tissues by combining two advanced techniques: the Gaussian Radial Basis Kernel Function and Spatial Neighborhood Information with a conventional fuzzy segmentation algorithm. The experimental findings indicate that the proposed method, integrating the Kernel Function and Spatial Information with Fuzzy Cluster Mean (SKFCM), is more effective in fibrosis detection than using either Kernelized Fuzzy Cluster Mean (KFCM) or Spatial Fuzzy Cluster Mean (SFCM) alone. Visual comparisons, segmentation validity metrics, and pixel accuracy data support this conclusion.

Ultrasound-based computer-aided diagnosis (CAD) has demonstrated considerable potential in identifying developmental dysplasia of the hip (DDH). More recently, foundation models have emerged as powerful tools for enhancing CAD performance. Despite these advances, applying such models in clinical environments is often limited by the substantial computational demands and the risk of overfitting associated with full model fine-tuning. To overcome these limitations, we introduce a memory-efficient adaptation strategy termed External Spatial Adapter Tuning (ESAT) for DDH diagnosis. ESAT incorporates spatial adapters constructed using low-rank depth wise separable convolutions at multiple layers of a foundation model, enabling effective extraction and refinement of hierarchical spatial features. The outputs of these adapters are externally aggregated for final classification, ensuring that gradient updates remain isolated from the foundation model parameters. Experimental evaluation on real-world DDH ultrasound datasets shows that ESAT consumes only 30.16% of the memory and 0.304% of the parameters required by full fine-tuning on a ViT-Base backbone, while achieving superior diagnostic accuracy compared with both full fine-tuning and existing parameter-efficient adaptation techniques.

The increasing penetration of renewable energy resources, rapid electrification and rising carbon emissions have transformed modern power systems into highly dynamic and vulnerable infrastructures. Conventional smart grid architectures primarily focus on reliability and economic operation, while limited attention has been devoted to integrated carbon-aware autonomous grid restoration. This paper proposes a novel Carbon-Aware Self-Healing Smart Grid (CASH-SG) framework using Digital Twin technology and Swarm Artificial Intelligence for intelligent fault prediction, autonomous reconfiguration and emission-aware energy optimization. The proposed framework creates a real-time virtual replica of the physical power network through a Digital Twin model that continuously monitors grid parameters, renewable intermittency, load variations and transmission contingencies. A hybrid Swarm AI optimization mechanism integrating Particle Swarm Optimization and Gravitational Search Algorithm is developed to minimize carbon emission, transmission loss and restoration time simultaneously during fault conditions. The framework dynamically identifies critical emission zones and performs adaptive feeder reconfiguration for resilient grid recovery. MATLAB/Simulink-based simulations are performed on IEEE standard bus systems under multiple fault and renewable uncertainty scenarios. Comparative analysis demonstrates that the proposed CASH-SG framework achieves superior carbon reduction, faster fault restoration, lower operational cost and enhanced system stability compared to conventional smart grid approaches. The proposed methodology introduces a next-generation intelligent energy management architecture suitable for future sustainable and autonomous power systems.

A noticeable part of the population suffers from temporary or permanent functional impairments due to disease or accidental injuries. In cases of difficult or impossible mobility, wheelchair becomes a need. But if we look at manual wheelchairs, they are helpful for low and medium level disabilities, and in severe cases it is difficult or impossible to use wheelchairs independently. As a result, individuals with severe disabilities often experience limited mobility and depend on others to go places with wheelchairs. To solve these issues, this research work aims at designing smart wheelchairs. Here we have explained how we made a wheelchair which has features voice control, joystick control, obstacle detection and SOS system. This wheelchair offers vital support for people with physical disabilities who lack constant surveillance. Users can operate this wheelchair using either joystick or microphone, providing them mobility regardless of the type of disability. Due to ultrasonic sensor, this wheelchair can detect obstacles ahead and alerts the user through a buzzer, signaling them to release the joystick or say "stop" via the microphone instantly. Additionally, our project presents a push button feature as well that sends alerts to the caretaker's mobile device in cases of danger.