Category Archives: Uncategorized

Authors: Utsav R. Hirapra

Abstract: As the scope of software projects increases, it becomes increasingly difficult to ensure proper code quality and conduct efficient code reviews, not because the required tools lack, but because all relevant information tends to become buried under unnecessary noise. In an attempt to combat that issue, CodeFox is presented as a lightweight, modular solution for discovering valuable insights by leveraging modern AI algorithms. In terms of functionality, CodeFox consists of a developer-friendly user interface, various automated review tools, and a meaning- based code search module. The platform collects metadata such as commits, reviews, comments, as well as ownership information and uses semantic embeddings to index critical elements within the code. By leveraging vector search algorithm, CodeFox allows its users to easily discover similar code, discussion threads related to the code, as well as reviewers who were working on this code. The use of the platform at an early stage of development in our company allowed us to shorten the review cycle process as well as reveal the reasoning behind code changes more efficiently. In this paper, we discuss CodeFox architecture, key aspects of integration, namely webhooks, background workers, and persistent storage, as well as share our experience of implementing a convenient yet lightweight platform to facilitate further development. We plan to conduct a user study in order to evaluate efficiency in the context of the problem and implement more advanced data gathering features and intelligent suggestions in the future.

DOI: https://doi.org/10.5281/zenodo.19603725

Published by: vikaspatanker

Authors: Aayush Yogesh Sanklecha, Pranav Dattatray Gund, Aditya Yogesh Salunke, Samarth Pramod Koli, Prof Mr.H.B.Gadekar

Abstract: Data quality remains the critical bottleneck in enterprise machine learning pipelines. Unreliable, schema-broken, drifted, or regulatory non-compliant data causes downstream analytics failures with consequences ranging from inaccurate predictions to regulatory penalties. This paper presents ADAP (Automated Data Analytics Platform) (Data Intelligence Pipeline with Expert Verification), an end-to-end data intelligence platform that unifies multi-source data ingestion, NLP-augmented semantic schema classification, seven-dimensional parallel validation, regulatory compliance enforcement (AML, HIPAA, SOX, GDPR), AutoML with SHAP explainability, and a dual reinforcement learning (RL) adaptation engine — all within a single auditable medallion-architected system. ADAP (Automated Data Analytics Platform) achieves schema classification accuracy of 94.7% across 31 semantic types, anomaly detection AUROC of 0.961, calibrated confidence scoring with ECE 0.0225 and AUC 0.9784, and multivariate drift detection at 89.4% accuracy at moderate distributional shift. A PPO Actor-Critic agent pre-trained over 1,000 synthetic episodes and warm-started via Thompson Sampling adapts 8-axis pipeline execution strategies in real time. End-to-end pipeline latency is under 7.4 seconds for 100,000-row datasets. All six production models pass tightened v7 quality gates, with performance validated end-to-end on held-out enterprise datasets.

Published by: vikaspatanker

Authors: Atchaya K, Madhumitha T, Pasumarthini R, Siva Sandhiya M, Susmitha S

Abstract: The rapid growth of urbanization has resulted in increased vehicle density on roads, raising the demand for efficient and intelligent traffic monitoring systems. This paper presents a real-time vehicle detection, tracking, and recognition system using YOLOv26(ultralytics), the latest advancement in the You Only Look Once (YOLO) architecture. The proposed system leverages deep learning-based object detection to detect and classify vehicles from video streams captured by surveillance cameras. YOLOv26(ultralytics) offers improved accuracy and speed over its predecessors, making it highly suitable for real-time Intelligent Transportation System (ITS) applications. The system incorporates Deep SORT for robust multi-object tracking and supports recognition based on vehicle attributes including color, type, and license plate.

Published by: vikaspatanker

Authors: Parag Hossain

Abstract: Traffic Sign Recognition (TSR) is a critical component of autonomous driving systems and Advanced Driver Assistance Systems (ADAS). However, real-world environmental challenges such as occlusions, lighting variations, multi-scale changes, and motion blur often degrade the performance of traditional vision pipelines. This paper presents a robust real-time TSR system based on the YOLOv7 architecture. The proposed model leverages an E-ELAN backbone for hierarchical feature extraction, a PANet neck for multi-scale semantic fusion, and an anchor-free IDetect head for precise localization. Trained on a custom traffic_sign_data dataset with aggressive data augmentation, the system achieves 94% mAP@0.5 and 38% mAP@0.5:0.95 at 45 frames per second on an NVIDIA RTX 3090 GPU. Comparative evaluations show that YOLOv7 significantly outperforms YOLOv5 with 89% mAP@0.5 and Faster R-CNN with 91% mAP@0.5 at only 12 FPS. The model is further optimized via ONNX-to-TensorRT conversion, enabling efficient deployment on edge computing platforms such as NVIDIA Jetson AGX Xavier.

DOI: https://doi.org/10.5281/zenodo.19594864

 

Published by: vikaspatanker

Authors: Biju Balakrishnan, Abdul Aziz Khatri, Hemang Korane, Yeshang Upadhyay, Eyakramul Hussain, Akshay Nugurwar

Abstract: This project provides a hybrid solution for the automated generation of timetables through the combination of Linear Programming (LP) and Constraint Satisfaction Problem (CSP) solutions. It’s intended to address the NP-hard problem of university course timetabling, which involves the satisfaction of a considerable number of conflicting constraints and preferences. The method for solving this problem involves two distinct steps. In the first step, a linear programming method is employed to determine the optimal assignment of teachers to subjects. This is a global optimisation technique that aims to maximize the satisfaction of teacher preferences and ensure a well-balanced allocation of teaching loads for all faculty members. In the second phase, a CSP algorithm with backtracking is applied to these optimal assignments to further assign time slots and classrooms. This phase is handled by both hard and soft constraints. Hard constraints, such as the unavailability of a teacher or a classroom, are represented by hard constraints that should not be violated to avoid any scheduling conflicts. Soft constraints, such as teacher time slot preferences and constraints on the lecture interval, are employed to filter the most optimal assignments from the pool of feasible solutions to enhance the quality of the schedule.

DOI: https://doi.org/10.5281/zenodo.19594429

Published by: vikaspatanker

Authors: Dr. S. Sharon Priya, Anish Fathima N, Blessy Evangeline A

Abstract: The campus environment in universities includes many blocks laboratories offices hostels libraries and other facilities spread across different areas which makes navigation difficult for students visitors and staff. The existing methods such as signboards printed maps and verbal directions do not provide proper support and users often struggle to find the correct location. The absence of real time guidance creates confusion especially during busy periods like admissions examinations and events where many people move at the same time. The difficulty increases when users are not familiar with the campus layout and this leads to time loss and frustration. The need for a simple reliable and easily accessible navigation system becomes very important in such situations. The proposed system is an AR based smart campus navigation solution that works through a mobile browser without the need for any application installation. The system uses GPS and device sensors to continuously track user position and direction and provides accurate navigation support. The A star algorithm is used to calculate the shortest path between the current location and the destination and this improves navigation efficiency. The guidance is shown using augmented reality arrows on the camera view which makes directions easy to understand and follow. The system also includes an AI chatbot for user queries voice guidance for hands free navigation and crowd prediction to avoid busy areas. The Firebase database is used to provide real time information such as lab availability staff details and event updates. The system provides a complete and user friendly solution for campus navigation.

Published by: vikaspatanker

Authors: Nitish Prajapati, Mahfooz Khan, Piyush Bagadi, Suraj Kumar

Abstract: In order to build solid relationships between colleges, their alumni, and present students, alumni en- gagement is essential. A specialized platform called ALUMNI CONNECT: BRIDGING THE GENERATION was created to strengthen these connections by offering a controlled, engaging envi- ronment for networking, career development, and mentoring. Key elements of the platform include job advertising, mentorship programs, discussion boards, guidance requests, regional alumni filters, alumni contributions, including donations, lab installations, and event planning. Furthermore, it maintains a trustworthy community by removing phony accounts and verifying profiles to guarantee authentic- ity. The initiative’s main driving forces are the underutilization of alumni networks and the dearth of easily accessible, individualized student counseling. By tackling these issues, the platform encour- ages deep connections, allowing alumni to commemorate successes and offer insightful career guidance and university-specific recommendations. Students can close the gap between academic accomplish- ment and career success by gaining access to industry connections and mentorship possibilities through increased engagement. In order to improve professional development and institutional linkages,this project seeks to establish a smooth and effective alumni-student engagement paradigm. ALUMNI CONNECT hopes to create a thriving self-sustaining ecosystem that benefits both students and alumni by utilizing technology to promote involvement.

Published by: vikaspatanker

Authors: Dr. Nilesh Jain, Kishan Vyas, Monil Lalwani, Kushal Shah

Abstract: The scope of enterprise reporting systems has grown significantly as far as data size, dashboard interactiveness, and visual analytics are concerned. However, in most companies, these systems operate as a static support for making decisions instead of an active decision system. The issue is that even though the dashboards detect patterns, anomalies, and trends, the user should understand these results and trigger the next steps. This implies some time lag, inconsistencies, and dependency on human decision-making. Thus, in this article, we propose a mixed AI-agent framework of enterprise decision systems where the intelligence ability of artificial intelligence is complemented by the execution capabilities of intelligent agents through staged development. The framework involves data collecting, cleaning, transformation, creation of dashboards, analysis with the help of machine learning algorithms for predictions, classifications, clustering, and detecting anomalies, and finally, the use of agents that translate model outputs into workflow activities such as approval, escalating, prioritizing, and notifying. The primary idea of this research is to split the process of intelligence generation and actions taking at different process stages, thereby increasing the modularity, interpretability, and effectiveness. However, it retains its practical applicability and relevance of dashboard analysis. The paper discusses the conceptual framework, technical design, model types, logic of process mapping, criteria of assessment, governance issues, and the possibility of enterprise implementation of the proposed approach

Published by: vikaspatanker

Authors: Chetan Uddhav Wagh, Yash Tatyaba Khandare, Shreya Prashant Mhaskar, Prof. K. R. Metha

Abstract: Artificial Intelligence based conversational systems are transforming how information services are delivered in many fields including education. Students frequently require quick access to academic information such as course details, admission procedures, faculty information, and examination schedules. However, traditional communication channels such as notice boards, websites, and help desks often fail to provide instant responses. This research proposes ChatGMVIT, an AI-powered chatbot designed to assist students by providing instant responses to institutional queries. The system is implemented as an Android application integrated with Firebase Firestore as a knowledge database and the Gemini 2.5 Flash Lite model for natural language processing.

DOI: https://doi.org/10.5281/zenodo.19593570

Published by: vikaspatanker

Authors: Mr.B. Janu Naik, Hemadri Sumedha, Anala Sanghavi, Giduthuri Charanchandu, Sathi Sanjana Reddy, Geetha Yeswanth Kumar

 

 

Abstract: Forecasting financial market trends has become an important task for investors, financial institutions, and analysts due to the increasing complexity and volatility of modern financial systems. Accurate prediction of market movements such as stock prices, exchange rates, and commodity prices can significantly assist in making informed investment decisions and managing financial risks. However, financial market data is highly dynamic, nonlinear, and influenced by multiple economic and external factors, which makes accurate forecasting a challenging problem for traditional statistical methods. In this study, a machine learning-based framework is proposed for forecasting financial market trends using time series analysis. The proposed approach utilizes historical financial data including stock prices, trading volumes, and other relevant financial indicators to train predictive models capable of identifying patterns and relationships within the data. A Support Vector Machine (SVM) algorithm is employed as the primary forecasting model due to its strong capability to handle nonlinear relationships and high-dimensional datasets effectively. The system performs several important stages including data loading, preprocessing, feature selection, model training, and performance evaluation. During preprocessing, missing values and irregularities in the dataset are handled, and normalization techniques are applied to ensure consistent feature scales. The processed data is then used to train the SVM model, which learns complex patterns present in historical financial data to generate predictions for future market trends. The performance of the forecasting model is evaluated using standard evaluation metrics such as Mean Squared Error (MSE) and Mean Absolute Error (MAE) to measure prediction accuracy. Experimental analysis demonstrates that the proposed machine learning framework is capable of effectively capturing nonlinear patterns present in financial time series data and providing reliable forecasting results. By leveraging machine learning techniques, the proposed system improves prediction efficiency and supports intelligent decision-making for traders, investors, and financial analysts. The framework can serve as a valuable tool for financial market analysis and can be further enhanced by integrating hybrid machine learning models and real-time financial data sources.

DOI:

 

 

Published by: vikaspatanker