Authors: Mukul Rane, Om Baviskar, Devendra Nikam, Tejaswi Malode, Associate Professor Vaibhav Dabhade

Abstract: This paper proposes DocInsight, a context-aware document analysis system that integrates preprocessing, Optical Character Recognition (OCR), layout analysis, and semantic processing into a unified pipeline. The system enhances text extraction accuracy while preserving document structure, en- abling efficient understanding of unstructured documents. By leveraging layout-aware OCR and transformer-based semantic models, DocInsight supports intelligent search, context-driven retrieval, and automated report generation. The framework ensures improved accuracy, structural consistency, and reduced manual effort in document processing. The system is applicable across multiple domains such as healthcare, legal systems, educa- tion, and enterprise environments, where efficient and intelligent document understanding is essential

DOI: https://doi.org/10.5281/zenodo.20743556

Published by: vikaspatanker

Authors: Miss. Urvi Pawar, Miss. Shreya Rakibe, Miss. Vaishnavi Sandhan, Miss. Samruddhi Vispute, Assistant Professor Dr. Kirti Patil

Abstract: Communicating with Deaf or hard-of- hearing people can be difficult at times and the chal- lenges are great in multilingual countries such as India. This study describes current work on this but offers a proposal for an animated ISL (Indian Sign Language) translation system from Marathi text and/or speech – based on the fact that there are many more English- MSL (Marathi Sign Language) resources available and, therefore, a ’significant access gap’ when considering Deaf users within our target country. The majority of currently available translation systems have been based upon machine learning; however, due to insuffi- cient parallel corpora/annotated sign data resources for Marathi-MSL, this method will not work. The proposed system adopts as an alternative a ’rule based’ method- ology which will map the Marathi language & grammar structures onto ISL using linguistic ’rules’ and dictio- nary and will develop this through web application using React.jsTailwindCSSFlask for the front end of the web application, while allowing use of ’browser based’ storage thus ensuring a very lightweight deployment. Given that ISL requires ’gestures’, ’facial expressions’ and ’spatial syntax’, it is not possible to translate word for word; rather, the system will also consider some important Marathi grammatical elements, e.g. inflection; location/post position, verb forms, sentence structure and will therefore generate an ISL output that more accurately represents the original Marathi and promotes communication and accessibility for Deaf individuals.

DOI: https://doi.org/10.5281/zenodo.20743457

Published by: vikaspatanker

Authors: Manthan Gavali, Om Malode, Shreeyash Jadhav, Yash Chaudhari, Assistant Professor Vaibhav Dabhade

Abstract: This project addresses the challenge of robust audio content identification in noisy environments by developing an ML-based audio fingerprinting system.To overcome this limi-tation, our methodology leverages a deep learning approach, using a Convolutional Neural Network to automatically extract a compact, noise-invariant fingerprint from audio spectrograms. The system involves a multi-stage process: a diverse dataset of clean audio is first augmented with various types of noise which has different signal to noise ratios. The trained model then generates a unique fingerprint for each audio track in a database. Finally, these fingerprints are stored using a fast and efficient hashing mechanism, enabling quick retrieval and identification. Our evaluation will demonstrate that this ML-based system significantly outperforms Existing methods in terms of accuracy and robustness, particularly at low SNRs, thereby providing a more reliable solution for applications such as music recognition, broadcast monitoring, and copyright enforcement.It further introduces spectrogram normalization and data-driven feature learning that minimize the impact of background dis-tortions. A contrastive-learning objective enforces the noisy and clean versions of the same audio to have similar embeddings. To facilitate fast retrieval, the system uses an approximate nearest-neighbor search mechanism optimized for large-scale databases. The approach’s low cost computational for fingerprint generation and matching is also demonstrated by experimental results. In general, the proposed approach allows for a scalable, high-performance framework suitable for real-time audio identifi-cation in adverse acoustic environments.This paper proposes a machine learning-based audio fingerprinting system for accurate audio identification in noisy conditions. A Convolutional Neural Network (CNN) is employed to learn noise-robust and compact audio fingerprints from audio spectrograms. Noise is added to clean audio examples with varying signal-to-noise ratio (SNR) values to enhance robustness. Contrastive learning is employed to guarantee that embeddings of noisy and clean audio examples are similar. The produced audio fingerprints are stored through a hashing function, and an approximate nearest neighbor search is employed for efficient retrieval. Experimental results show enhanced audio identification accuracy with low computational complexity in low SNR conditions. The proposed system is appropriate for scalable and real-time audio identification tasks

DOI: https://doi.org/10.5281/zenodo.20743337

Published by: vikaspatanker

Authors: Ishaan Puri, Assistant Professor Dr. Raghavendra Barshikar

Abstract: Ball valves are popular in factories for controlling liquids and gases because they are easy to make, last a long time, and shut off flow well. But regular ball valves are not always great at controlling how much liquid or gas flows through, especially when they are only partly open. V-port ball valves were created to fix this problem. The V-shaped cut in the ball helps to control the flow better. This research looks at how a V-port ball valve works using computer simulations. A 3D model of the valve was made, and simulations were run with different opening amounts and flow rates. The research looked at things like how fast the liquid or gas moves, how much the pressure drops, how much turbulence there is, and the valve's flow rate. The results show that the V-port design does a much better job of controlling flow than regular ball valves. It makes the flow smoother and reduces the chance of bubbles forming. The results also show that computer simulations are helpful for making better valve designs. The simu

DOI: https://doi.org/10.5281/zenodo.20743255

Published by: vikaspatanker

Authors: Gaurav A. Bagul, Parth P. Jadhav, Pratik S. Rahane, Assistant Professor Vipin K. Wani

Abstract: Cryptocurrencies have rapidly grown into a pop- ular medium of digital exchange, offering speed, security, and borderless transactions. While these benefits have driven global adoption, the pseudony- mous and decentralized nature of cryptocurrencies also makes them highly vulnerable to misuse in ille- gal activities such as money laundering, terrorism financing, and fraud. Recent reports highlight bil- lions of dollars being laundered annually through cryptocurrency channels, often using techniques like mixers, peel chains, and cross-chain transfers. Traditional Anti-Money Laundering (AML) sys- tems, designed mainly for conventional banking transactions, struggle to handle the complexities of blockchain-based transactions. They often func- tion as black boxes, providing risk scores without clear reasoning, and they are reactive rather than proactive in detecting suspicious activities. To address these challenges, the proposed sys- tem CryptoTrack: A Data-Driven System for De- tecting Cryptocurrency Laundering. The system leverages advanced analytics to identify suspicious accounts and transactions, while integrating Ex- plainable Artificial Intelligence (XAI) to provide transparent justifications for every detection. Un- like existing systems that only flag activities, Cryp- toTrack enables users and compliance officers to understand the exact reasons why a transaction is considered risky, thereby increasing trust and re- ducing false positives. A visualization dashboard further supports users by providing intuitive in- sights into detected suspicious activity. The proposed framework bridges the gap be- tween opaque detection models and the practical requirement for interpretability in financial mon- itoring. By combining data-driven detection, ex- plainability, and transparency, CryptoTrack offers a more reliable and effective approach to combating financial crimes in the rapidly evolving landscape of cryptocurrency.

DOI: https://doi.org/10.5281/zenodo.20743147

Published by: vikaspatanker

Authors: Reena

Abstract: The present investigation was carried out to evaluate the effect of different agricultural substrates on the quality production of oyster mushroom (Pleurotus sajor-caju #392). The study was conducted using four agricultural waste substrates, namely wheat straw, rice straw, sugarcane bagasse, and maize straw for mushroom cultivation. The experiment focused on spawn running, pinhead appearance, maturity, flush-wise yield, biological efficiency, and nutritional composition of cultivated mushrooms. Results revealed that wheat straw showed the fastest spawn running and pinhead formation with maximum total yield and biological efficiency. Wheat straw recorded 1360 g total yield and 136% biological efficiency, while sugarcane bagasse showed the lowest yield performance. Nutritional analysis indicated that sugarcane bagasse had the highest protein content (8.75%), whereas maize straw recorded maximum crude fat content (10%). Wheat straw exhibited superior fiber and ash content. The findings conclude that wheat straw is the most suitable substrate for commercial cultivation of Pleurotus sajor-caju, whereas sugarcane bagasse can be used for improving mushroom nutritional quality. This study highlights the importance of selecting appropriate substrates for achieving better mushroom production and quality.

DOI: https://doi.org/10.5281/zenodo.20742909

Published by: vikaspatanker

Authors: Gayatri Pandram

Abstract: Endophytic fungi are ubiquitous microorganisms that asymptomatically colonize the internal tissues of plants, representing an untapped reservoir of novel, biologically active secondary metabolites. This study investigates the endophytic fungal diversity associated with two ethnobotanically critical medicinal plants: Withania somnifera (Ashwagandha) and Amomum subulatum (Badi Elaichi), and evaluates their biomedical potential against clinically significant human pathogens. Healthy leaves, stems, and roots were subjected to a stringent multi-step surface sterilization protocol and inoculated onto Potato Dextrose Agar (PDA). A total of fifteen (15) distinct fungal endophytes were isolated and taxonomically characterized via macroscopic and microscopic morphotyping. The predominant genera identified included Aspergillus, Fusarium, Alternaria, Curvularia, Penicillium, and Phoma. The cell-free secondary metabolites were extracted using organic solvents and screened for antibacterial efficacy against Klebsiella pneumoniae, Staphylococcus aureus, Escherichia coli, Salmonella typhi, and Micrococcus spp. using the agar well diffusion assay. The bioprospecting profile revealed significant inter-species variability. Notably, Aspergillus flavus derived from W. somnifera exhibited a profound, broad-spectrum zone of inhibition against both Gram-positive and Gram-negative cohorts, with optimal metabolic yield quantified at a 7-day incubation kinetics threshold. These insights underscore the therapeutic relevance of plant-associated fractions as sustainable alternatives to combat escalating multi-drug resistance (MDR) phenotypes.

DOI: https://doi.org/10.5281/zenodo.20742728

Published by: vikaspatanker

Authors: Research Scholar Akhilesh Singh, Associate Professor Dr Namita Tiwari, Associate Professor Dr Mayur Rahul

Abstract: With an increasingly large number of online services and secure access applications, trusted identity authentication has become an important issue. Although biometric authentication has higher security assurance than traditional security methods, single biometric mode authentication systems have performance issues in terms of degradation due to environmental factors, occlusions, lighting, and spoofing attacks. In this respect, this study proposes an original multimodal biometric authentication approach that combines ear contour biometric recognition with palmprint biometric recognition using the Enhanced and Discriminative Competitive Code (EDCC) method. The proposed multimodal biometric authentication method has the synergistic ability of two biometric modes. The ear contour-based biometric recognition technique extracts the helix and conchal curvatures of the human ear, providing geometric information that is less affected by illumination conditions. Simultaneously, the EDCC-based palmprint recognition technique extracts the prevalent orientation patterns of the lines and ridges on the human palm, providing robustness to noise and minute geometric distortions. These two biometric modalities provide complementary information about the user’s biometric traits and can thus be fused through feature-level fusion to provide a single and robust biometric representation.The performance of the proposed multimodal biometric authentication technique is evaluated on two challenging and widely available biometric datasets, namely the PolyU-IITD contactless palmprint database and the EarVN1.0 unconstrained ear image database. The performance evaluation of the proposed multimodal biometric authentication technique clearly reveals its superior performance compared to other state-of-the-art biometric authentication approaches, including unimodal and hybrid biometric authentication schemes, as it provides a recognition accuracy of 99.01% and an extremely low EER of 0.11% for the PolyU-IITD contactless palmprint database and EarVN1.0 unconstrained ear image database, respectively.

DOI: https://doi.org/10.5281/zenodo.20742447

Published by: vikaspatanker

Authors: Daksha Tigga

Abstract: Drosophila melanogaster serves as an indispensable in vivo model system for mapping how external environmental variations and nutritional inputs dictate physiological adaptations and behavioural choices. This study presents a multi-generational evaluation of how dietary variance (standard cornmeal vs. banana-enriched vs. orange-enriched media) pairs with photoperiodic conditions to modulate ontogeny, locomotor agility, and larval chemotaxis. Across three successive generations (F1–F3), cohorts reared on a nutrient-dense banana medium exhibited accelerated metamorphic transitions and robust pupation rates. Conversely, an orange-supplemented diet delayed developmental milestones and reduced total yield compared to uniform controls. Photoperiodic restrictions (sustained dark phases) consistently decelerated growth metrics and decreased motor output across all dietary groups. Quantifiable behavioural deficits under low-light regimes were verified via negative geotaxis assays, where light-exposed flies displayed markedly superior vertical climbing performance. Furthermore, larval olfactory assays revealed a stark chemotactic bias toward volatile food-derived attractants (ethyl acetate) over aversive ionic stimuli (sodium chloride). Taken together, these data illuminate the complex interplay between systemic metabolic programming and sensory-driven behavioural phenotypes in response to immediate ecosystem shifts.

DOI: http://doi.org/10.5281/zenodo.20741467

Published by: vikaspatanker

Authors: Professor Sonali Dongare, Priyanshu Singh, Aditya Amup

Abstract: Motor impairments such as ALS, locked-in syndrome, and cerebral palsy severely limit an individual's ability to interact with digital systems using conventional input devices. This paper presents GazeSpeak, an AI-powered Eye Gaze Communication System that enables motor-impaired users to communicate through voluntary eye movements alone. The system extracts real-time gaze coordinates using OpenCV and MediaPipe, maps them onto interactive screen elements via a TensorFlow regression model, and integrates a transformer based NLP module for context-aware word prediction. A dwell-based selection mechanism activates interface targets without any physical input. Experimental evaluation across twenty participants demonstrates a gaze detection accuracy of 94.2%, end-to-end latency of 38ms, top-3 word prediction accuracy of 87.6%, and communication throughput of 10.6 WPM, with a System Usability Scale score of 84.4 confirming excellent user acceptance. The results establish GazeSpeak as an effective, open-source, and cost-accessible assistive communication platform for real-world deployment.

DOI: https://doi.org/10.5281/zenodo.20735406

Published by: vikaspatanker