Welcome to my GitHub profile! I'm a passionate software developer and data enthusiast with a keen interest in machine learning, natural language processing, and neural networks. My journey in technology began with a curiosity about how data can be harnessed to solve real-world problems, leading me to dive deep into the realms of AI and data science.

From tinkering with simple Python scripts during my college days to building complex machine learning models, my journey in coding began in a scientific environment—specifically, physics. In this role, I encountered image processing, which sparked a deep dive into the field. Through this exploration, in a very short time, I mastered technologies like CUDA and Vision Transformers. Seeking to streamline our workflow, I implemented a straight-off-the-shelf Vision Transformer, which was initially overkill for the project's needs. However, I was astonished by the model's performance, and this experience left me mesmerized by the limitless possibilities of AI.

This pivotal moment led me to pursue a master's degree in AI, where I was exposed to a variety of intriguing niches within artificial intelligence. I've delved into each of these areas and remain passionate about AI as a whole, with a particular focus on Machine Learning.

These niches have been both fascinating and challenging. Each of the following projects required substantial effort, often spanning months or years. Despite the challenges, each endeavor has been a rewarding learning experience, strengthening my problem-solving abilities and technical expertise. I am committed to continuing this journey, eager to take on new challenges and contribute to innovative solutions across various fields, as demonstrated by my projects below.

A comprehensive collection of projects focused on Natural Language Understanding (NLU). Utilizing well-established statistical methods, this repository covers tasks such as entity recognition, language modeling, and syntactic parsing, highlighting the depth and breadth of my work in NLP.

This project builds a Sentiment Analysis model using Convolutional Neural Networks (CNNs). It tackles the challenges of representing words and phrases as vectors, enabling the model to accurately discern and categorize sentiments from textual data.

In Progress: The goal of this project is to fully train a reproduction of GPT-3 which is just a few generations behind current model used by OpenAI. This Project aims to understand the nitty-gritty of LLMs and gaining a more nuanced understanding of certain features/bugs of LLMs including but not limited to Hallucinations. Due to hardware limitations, we will train a smaller model but the core of the model should work identically.

A Layer 7 Load Balancer Dashboard project featuring a high-performance C backend implementing the Weighted Least Connections algorithm and a user-friendly Flask frontend. It includes RESTful APIs for managing load balancing rules and backend servers, real-time metrics monitoring, configuration management, and comprehensive logging, ensuring efficient traffic distribution and seamless frontend-backend communication. This showcases my proficiency building full scale applications and in-depth software understanding.

In this project, I delved into State Space Models to perform statistical approximations of complex time series data. Faced with unpredictable factors like human behavior and volatile weather patterns, I developed a robust model that provides accurate predictions amidst uncertainty, demonstrating the practical applications of statistical methods in real-world scenarios.

Utilizing Support Vector Machines (SVM), this project performs numerical approximation of time series data. Perfect for datasets influenced by predictable factors like natural phenomena or consistent physical processes, I explored how SVMs can effectively capture underlying patterns to deliver precise forecasts.

Leveraging Long Short-Term Memory (LSTM) Networks, this deep learning project models recurrent processes within time series data. Tailored for stock market predictions, it captures the intricate dependencies and recursive changes in stock prices, offering highly accurate forecasting capabilities.

An in-depth exploration of Reinforcement Learning techniques, this project implements both DQN and A3C methodologies. It delves into advanced topics like Transfer Learning, Trust Region Policy Optimization (TRPO), and Proximal Policy Optimization (PPO), providing a comprehensive survey and practical implementations.

A critical analysis and implementation of the groundbreaking Transformers architecture introduced in the seminal paper "Attention is All You Need." This project covers various Transformer applications, particularly in Natural Language Processing (NLP), demonstrating their transformative impact on AI.

This project delves into Contrastive Learning, a self-supervised learning paradigm. It examines different architectural approaches and loss functions, highlighting how contrastive methods enhance feature representation without extensive labeled data.

Exploring Diffusion Models with a learned encoder, this project focuses on image generation and other generative tasks. It showcases how diffusion processes can be harnessed for high-quality generative purposes, pushing the boundaries of what's possible in generative AI.

An implementation of Faster R-CNN, this project enhances traditional CNN models by introducing a more efficient processing pipeline. The result is a model that not only speeds up computations but also maintains, and even improves, accuracy in object detection tasks.

This project implements Variational AutoEncoders (VAEs), a type of generative model designed to create new images based on training datasets. By training on diverse image sets, the VAE learns to generate realistic and varied images, showcasing the power of encoder-decoder architectures in image synthesis.

A hands-on project with Deep Convolutional GANs, this repository experiments with generating handwritten digits from the MNIST dataset. By ingesting random noise, the GAN learns to produce authentic-looking handwritten numbers, demonstrating the effectiveness of adversarial training in generative models.

Aiming to develop a real-time Noise Cancellation model, this project initially started with an AutoEncoder architecture to remove noise from data streams. Facing architectural challenges, it pivoted to a 1D U-Net model, achieving comparable results with significantly reduced computational time. Explore the journey and the solutions implemented to enhance real-time noise reduction.

This project automates the design of deep learning models using Neural Architecture Search (NAS). Leveraging both Reinforcement Learning (RL) and Evolutionary Algorithms (EA), it dynamically generates and optimizes architectures for image classification on the CIFAR-10 dataset. The project explores how NAS can outperform manual architecture design by efficiently balancing accuracy and computational cost, based on insights from the paper "Neural Architecture Search: Insights from 1000 Papers." The final models are evaluated based on accuracy, complexity, and efficiency.

In Progress (prototype ready): This project focuses on Image Compression using neural networks. By leveraging advanced compression algorithms, the goal is to reduce image sizes without compromising quality, paving the way for efficient storage and transmission solutions.

Currently, I'm focused on advancing my expertise in Machine Learning. My goal is to develop scalable ML solutions that can transform industries by enhancing efficiency and unlocking new possibilities. I'm also keen on contributing to open-source projects and collaborating with like-minded professionals to drive technological innovation.

• Programming Languages: Python, CUDA, C/C++, JavaScript
• Frameworks & Libraries: TensorFlow, PyTorch, Keras, scikit-learn, REST APIs
• Tools & Technologies: Git, Jupyter Notebooks, Supercomputers
• Areas of Expertise: Machine Learning, Deep Learning, Software Development, Natural Language Processing, Data Analysis, Time Series Forecasting

• Tech Enthusiast: Always exploring the latest advancements in AI and machine learning by keeping myself updated on the arxiv and religiously reading new ML papers.
• Lifelong Learner: Committed to continuous education through courses, workshops, and hands-on projects. I started my journey in Softare Dev as a self-learner so, my drive to keep learning new ideas is unparalleled.
• Hobbyist Photographer: Enjoy capturing moments and experimenting with image processing techniques, started producing time-lapses using cameras and openCV in High School, still do it from time to time.

I'm always excited to connect with fellow developers, researchers, and enthusiasts. Whether you're interested in collaboration, have questions about my projects, or simply want to chat about the latest in AI, feel free to reach out!

• LinkedIn
• Email

Thank you for visiting my profile! 😊 Feel free to explore my repositories and reach out if you'd like to collaborate or discuss ideas.