Arjun Ramesh Kaushik Email & Phone Number

Buffalo, Ny, Us

Buffalo, New York, United States

- Image Generation : Working on generating images from brain waves (EEG / MEG) using an alignment module and diffusion models.- Vision-Language Models (Sports Analytics) : Ongoing work on developing a multi-modal model for real-time commentary generation (dense video captioning) in soccer videos using a Retrieval Augmented Generation (RAG) architecture and Knowledge Graphs. - Face Analytics / Face Recognition : Developed a biometric protection template for face embeddings using FHE and multivariate polynomial transformation. Template only allows identity verification and protects soft-biometric features. Extended the work by using Matryoksha Representation Learning to compress embeddings before encryption.- Large Language Models : Working on discerning human-authored text from AI-authored text through embedding fusion of different Language Models. Achieved high levels of classification accuracy on different datasets including Human vs LLM and SQuAD.- Signal Processing : Developed an end-to-end secure 1D CNN classifier network in FHE to detect Sleep Apnea. Leveraged Homomorphic Fourier Transform (HFT) that uses Cooley-Tuckey matrix factorization to perform 1D DFT of the input signal, improving processing speed by 170x. - Autonomous Unmanned Aerial Vehicles (UAVs) : Constructed an CNN-Actor-Critic Network on encrypted inputs, improving privacy in drone navigation. Derived polynomial approximations of non-linear functions through Polynomial Regression. Improved processing speed by 50% through parallelization. Extended work through Knowledge Distillation to reduce processing time by 18x.- Graph Neural Networks : Worked on Edge Convolution and Conv-LSTMs for encrypted data to construct the Graph Convolutional Neural Networks pipeline for fMRI analysis. Improved processing speed by 30% through parallelization.

Troy, Michigan, United States

- Designed, developed, and deployed a multi-modal physical-task assistance system. Generates textual and visual instructions based on user query and additional context from Knowledge DB (through RAG) to help solve industrial tasks. Building blocks – Llama3.1, LLaVA, CLIP, Whisper, Streamlit.- Contributed to a repository-level CodeLLM framework to understand large codebases and resolve bugs.- Contributed to a foundational model for semantic segmentation of defects in car seats.

- Operated as an independent full-stack developer in a remote setting, focusing on enhancing the functionality and appearance of the company's website.- Crafted comprehensive solutions using C#, AngularJS, Java, and SQL, demonstrating a strong analytical and problem-solving ability, which improved website functionality and user experience.- Established efficient CICD pipelines using tools such as Jenkins, Docker, and Terraform, leading to streamlined deployment of company-imported products in AWS and Microsoft Azure.

- Engaged in a remote research role within the Zhang Lab, focusing on the development and application of machine learning models to address complex data segregation and privacy concerns.- Pioneered a Centralized Federated Variational Autoencoder model using the Federated Averaging algorithm, facilitating the segregation of data into 5 distinct clusters based on 90635 features. This innovative approach yielded a silhouette coefficient of 0.855, demonstrating the model's effectiveness in data segregation. - Concurrently, addressed privacy concerns in Federated Learning by leveraging Differential Privacy, ensuring secure and private data handling.

Mumbai, Maharashtra, India

- Engaged in a dynamic 3-member team, innovatively applying ReactJS, Java, and MySQL to replicate the Monopoly board game into real-life banking scenarios, enhancing the user experience through the development of REST APIs.- Identified and resolved a critical issue of webpage crashing due to concurrent users, leading to improved system stability. Further optimized the testing process by formulating Cypress test-packs, automating manual testing and ensuring seamless user experience.

Chennai, Tamil Nadu

- Guided a team of seven in the development and implementation of a voice interactive web app for Rajiv Gandhi Government General Hospital, focusing on streamlining doctor appointments and patient guidance.- Programmed as a full-stack developer for end-to-end development of the web app, utilizing HTML, CSS, JavaScript, and Python, which included the integration of a chat-bot to enhance user interaction.- Deployed the web app using Firebase, ensuring accessibility and scalability of the platform, leading to improved user experience and interaction.

Bengaluru, Karnataka, India

Contributed to the development of a face recognition system. Computed Local Binary Patterns over face regions and utilized a Support Vector Machine for precise classification of genuine and counterfeit faces

Doctor Of Philosophy - Phd, Computer Science

Masters, Computer Science, 3.71 / 4.0

Coursework - Reinforcement Learning, Machine Learning, Computer Vision, Natural Language Processing, Deep Learning, Sports Analytics.

Minor, Robotics And Automation

Bachelor'S Degree, B.E. (Hons) Electronics And Instrumentation Engineering

Activities and Societies: Vice Captain - Soccer Team

Physics, Chemistry, Math And Electronics

9.8 Cgpa

Activities and Societies: Captain - Soccer Team, Cricket Team