SSSuthir Sriram

Suthir Sriram

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Suthir Sriram has a total of 35 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    RAG-NetTwin: A hallucination-mitigated framework for heterogeneous intent-based networking via graph-augmented retrieval and digital twin

    by Nivethitha Vijayaraj, Suthir Sriram, Aryan Madhusudhanan, Srivishnu T․, Gaurav Choudhary

  • MARINE-Net: Physics-guided lightweight CNN-INR architecture for underwater image restoration and color enhancement

    by Suthir Sriram, Nivethitha Vijayaraj, Swamynathan V.P., Gaurav Choudhary, Thangavel Murugan


  • Top Cited Papers
    Open Access

    A Reusable Prompting Framework for Applying Large Language Models to Legal Tasks

    by Suthir Sriram, Nivethitha Vijayaraj, G. Rajiv Krishna, S. Vijay Bhanu, Gaurav Choudhary, Thangavel Murugan

    Large language models are increasingly used for legal research, drafting, and analysis, but their answers can be inconsistent, hard to verify, and sensitive to small changes in prompt wording. This paper introduces a structured prompting framework for applying large language models to five core legal tasks: statutory interpretation, contract review, case summarization, legal question-answering, and clause extraction. The framework helps draft prompts using task-specific templates, role-based instructions, example-based guidance, chain-of-thought reasoning and context-layered reasoning, and is paired with an evaluation that measures exact match, F1, ROUGE-L, macro F1, and a legal hallucination rate defined through a rubric for unsupported or incorrect legal claims. To contextualize these results, a Retrieval-Augmented Generation baseline is implemented and evaluated on the same tasks. Experiments across multiple state-of-the-art models show that the structured prompting framework consistently improves accuracy and stability over less structured prompts and over the Retrieval-Augmented Generation baseline in this setting, with the largest gains for reasoning-intensive tasks such as statutory interpretation and legal question-answering. Context-layered and chain-of-thought prompts yield the highest scores and the lowest hallucination rates overall. Statistical analysis over multiple runs, using mean, variance, standard deviation, and 95% confidence intervals, indicates that the best model–prompt combinations are both strong and stable. The findings offer practical guidance on which structured prompting strategies are most effective for different categories of legal tasks.


  • Open Access

    An Adaptive YOLO Framework for Coral Reef Object Detection With DUAA, MLCA, and CAL

    by Suthir Sriram, Nivethitha Vijayaraj, R. Sriraam Vairavan, Yadhav Balaji Rao, Gaurav Choudhary

  • Open Access

    Advanced MRI based Alzheimer’s diagnosis through ensemble learning techniques

    by Suthir Sriram, V. Nivethitha, T. P. Arun Kaarthic, S. Archita, Thangavel Murugan

    Alzheimer’s Disease is a condition that affects the brain and causes changes in behavior and memory loss while making it hard to carry out tasks properly. It’s vital to spot the illness early, for effective treatment. MRI technology has advanced in detecting Alzheimer’s by using machine learning and deep learning models. These models use neural networks to analyze brain MRI results automatically and identify key indicators of Alzheimer’s disease. In this study, we used MRI data to train a CNN for diagnosing and categorizing the four stages of Alzheimer’s disease with deep learning techniques, offering significant advantages in identifying patterns in medical imaging for this neurodegenerative condition compared to using a CNN exclusively trained for this purpose. They evaluated ResNet50, InceptionResNetv2 as well as a CNN specifically trained for their study and found that combining the models led to highly accurate results. The accuracy rates for the trained CNN model stood at 90.76%, InceptionResNetv2 at 86.84%, and ResNet50 at 90.27%. In this trial run of the experiment conducted by combining all three models collaboratively resulted in an accuracy rate of 94.27% compared to the accuracy rates of each model working individually.


  • Open Access

    GPU Accelerated Trilateral Filter for MR Image Restoration

    by Suthir Sriram, Nivethitha Vijayaraj, T. Srilekha, M. Praveena, Thangavel Murugan

    Medical image processing demands essential image restoration techniques to handle both blurred and noisy images. The image capture process frequently causes these degradations. The restoration of medical images such as MRI scans holds essential value for better diagnosis and improved treatment accuracy. MRI serves as a non-invasive imaging tool to diagnose brain diseases yet the noise produced during its acquisition phase degrades image quality so that diagnosis accuracy suffers. Reduction of noise becomes essential for both clinical diagnostic procedures and computer-assisted medical analysis practices which require tissue classification and registration and segmentation algorithms. Detecting and eliminating noise from magnetic resonance images represents a complex medical imaging challenge. Multiple noise filtering methods operate on MR images with distinct performance benefits. This work presents a trilateral filter specifically designed for image restoration which provides both precision and speed performance. The bilateral filter served as foundation to develop this filter while adding another intensity comparison function to deal with brain MR image noise. A weighting control function based on intensity entropy is established for intensity-based weight calculations. The implementation of GPU parallel computing techniques together with optimization of memory and threads leads to faster computation. A combination of texture-based image analysis with advanced computational algorithms powers the automated filtration process. The approach uses forward selection to identify 98 texture attributes while refining the selection process to find optimal regularity features. A two-phase classification system trains automation parameters using artificial neural networks together with support vector machines. Research findings show that trilateral filtering yields superior noise reduction alongside better definition of MR image features than alternative techniques.


  • Open Access

    Anterior Cruciate Ligament (ACL) Tear Detection Using Hybrid CNN Transformer

    by Suthir Sriram, Deependra K. Singh, D. V. Sairam, Nivethitha Vijayaraj, Thangavel Murugan

    Classifying medical images to detect ACL (Anterior Cruciate Ligament) tears is the medical image classification task, which requires significant importance to avoid misdiagnosis. However, this also poses some difficulties due to differences in formats of medical images and standard images. While Convolutional neural networks (CNN) and transformers are among the types of models that have been used for such tasks, they both have their weaknesses. For instance, CNNs have a tendency of extracting local features, hence have problems in understanding global context but on the other hand, transformers excel in understanding global information even if it implies losing some details in localization. Because of this, we introduce the MambaConvT (Mamba Convolutional Transformer) model that employs a hybrid strategy that combines CNNs and transformers giving room for the advantages of both. Firstly, MambaConvT utilizes multi-core convolutional networks to achieve higher extraction capability of the ACL tear specific local features from structural MR (Magnetic Resonance) images. Next, it uses a selective scanning module to implement depth separable convolution using Selective Scanning Module (SS2D) to achieve a larger distance of the receptive field therefore increasing the chances of getting important long-distance connections and essential fine details for the tear detection process. A hybrid model is utilized that integrates these hybrid features and after global feature modeling, attempts to enhance both local and global information representation. MambaConvT outperforms the current major classification models, according to a robust performance evaluation conducted on four public and two protected datasets. It also yields the best detection scores, taking into account all evaluation metrics, such as accuracy, recall, precision, F1 score, and area under the curve, for the identification of ACL tears in medical imaging.


  • Open Access

    Enhancing Digital Identity and Access Control in Event Management Systems Using Sui Blockchain

    by Suthir Sriram, P. R. Tharaniesh, Priyanka Saraf, Nivethitha Vijayaraj, Thangavel Murugan

    The evolution of blockchain technology presents transformative potential for event management, particularly in enhancing security, transparency, and efficiency across the industry. This paper introduces a blockchain-based event management system that leverages Sui blockchain technology to address critical challenges such as ticket fraud, data security, and high transaction fees, which are prevalent in traditional event management systems. The proposed system incorporates several innovative features: Tickets to the System: Utilizing the concept of NFT, the system guarantees that each ticket is unique, is tamper proof and cannot be forged. It also ensures the validity of tickets and at the same time ensures that the ticket is transferable back and forth while the ownership of the ticket is wholly and clearly disclosed. zkLogin for Seamless User Authentication: zkLogin is then implemented in the platform to ensure users can be authenticated in a Privacy-Preserving manner while at the same time having High Security. This mechanism can be highly beneficial for the users because they can log in as quickly and discreetly as possible without any threat to their accounts’ safety. Suiet & Sui Wallet for Cost-Effective Transactions: Merging of Suiet and Sui Wallet enables cheap transactions and something that is always financially draining when using blockchain systems is deposits costly transactions fees. This makes the platform to be open to a wider market especially to those who are organizing small scale events. Due to decentralizing the event management process the platform reduces all possible risks that can occur in condition of centralized systems like for example single points of failure and also creates higher transparency of ticket sales and distribution. Furthermore, this greatly decentralizes the system which drastically decreases the transaction costs making the solution fiscally sustainable for events of various scales. This paper also presents the design and implementation of the platform as well as the flexibility of the platform depending on the kind and size of the event. The research indicates that the use of this blockchain technology increases the security and the verifiability of the event tickets in addition to providing a solution to the whole flow of the event beginning from ticket distribution to ticket authentication and transaction management.


  • Open Access

    Enhanced YOLOv10 Framework Featuring DPAM and DALSM for Real-Time Underwater Object Detection

    by Suthir Sriram, Aburvan P., Arun Kaarthic T. P., Nivethitha Vijayaraj, Thangavel Murugan

    Recently, spotting underwater objects has been increasingly difficult due to the complexities of marine environments and varied visibility conditions. YOLOv10 is notable for its effective, robust architecture, featuring significant components: advanced backbone networks and enhanced feature pyramid networks that also deliver anchor-free detection. In delivering YOLOv10, we enhance it with dual partial attention mechanism (DPAM) and dual adaptive label assignment with sun glint removal module (DALSM) along with marine fusion loss (MFL). With DPAM, the latest refinement processes for conservation focus on feature extraction to account for key highlights and in the scene, include temporal context, both critical for interpretation of the dynamic realm below the ocean. The prefix with DALSM involves adaptive dual label assignment and techniques for sun glint removal. The marine fusion loss (MFL) provides an object detection prediction that combines both binary cross-entropy loss and complete intersection over union (CIoU) loss to enhance bounding box localization while also including spatial context to incorporate important underwater features. With the experiments, we attenuate enhancements with device gradient clipping, model checkpointing, and advanced augmentation processes to gain 3.04% improvement in mean Average Precision (mAP). These enhancements mitigate perceived challenges of underwater detection while enhancing knowledge of understanding marine life.


  • Open Access

    Leveraging Local LLMs for Secure In-System Task Automation With Prompt-Based Agent Classification

    by Suthir Sriram, C. H. Karthikeya, K. P. Kishore Kumar, Nivethitha Vijayaraj, Thangavel Murugan

    Recent progress in the field of intelligence has led to the creation of powerful large language models (LLMs). While these models show promise in improving personal computing experiences concerns surrounding data privacy and security have hindered their integration with sensitive personal information. In this study, a new framework is proposed to merge LLMs with personal file systems, enabling intelligent data interaction while maintaining strict privacy safeguards. Our methodology organizes tasks based on LLM agents, which apply designated tags to the tasks before sending them to specific LLM modules. Every module is has its own function, including file search, document summarization, code interpretation, and general tasks, to make certain that all processing happens locally on the user’s device. Our findings indicate high accuracy across agents: classification agent managed to get an accuracy rating of 86%, document summarization reached a BERT score of 0.9243. The key point of this framework is that it splits the LLM system into modules, which enables future development by integrating new task-specific modules as required. Our findings suggest that integrating local LLMs can significantly improve interactions with file systems without compromising data privacy.