From a young age, I have always been fascinated by mathematics and technical concepts. Solving math problems felt like cracking puzzles, and understanding how things worked gave me a sense of wonder. This curiosity naturally guided me toward engineering, and earning a seat at IIT Kanpur was a dream come true, a chance to explore my interests at one of India's premier institutions.

My journey into Machine Learning began through IITK's vibrant Science and Technology clubs. During their recruitment tasks, I was introduced to the basics of ML and was immediately hooked. The ability to use data to make predictions, spot patterns, and solve problems felt like an exciting extension of my love for mathematics. Inspired by this experience, I decided to dive deeper into the field, and it didn't take long for opportunities to come my way.

First Real-World Project

Indian Railways My first major exposure to ML came through a project with Indian Railways, under the mentorship of Prof. Gururaj MV from the EE Department. It all began when Prof. Gururaj sent a mail to EE students about a project opportunity. Eager to apply what I had learned, I reached out to him, and soon I was working on one of the most impactful projects of my journey so far. The project involved predicting energy load usage for Indian Railways in Punjab, with the goal of reducing unnecessary costs and carbon emissions. It was my first taste of applying ML to solve real-world problems, and the stakes felt high. I worked on building an Al-driven time series analysis model, which required handling large datasets, preprocessing them for accuracy, and fine-tuning the model for optimal predictions. After several iterations and challenges, I developed a model that achieved 95% accuracy. Seeing my work contribute to a project that had real-world environmental and financial impact was both fulfilling and inspiring. It showed me the power of ML when applied to the right problems, and it motivated me to continue exploring the field.

Research Experiences

1. SURGE Program with Prof. Koteswar RJ: As I delved deeper into ML, I wanted to explore its theoretical side, which led me to pursue a research internship under the SURGE program. In early 2024, I began reaching out to professors whose work aligned with my interests. After several discussions, I connected with Prof. Koteswar RJ from the EE Department. His proposal incorporating complex-valued weights and biases into Fully Convolutional Neural Networks (FCNs) immediately intrigued me. It was ambitious, uncharted territory, and exactly the kind of challenge I was looking for. The project was anything but straightforward. The first hurdle was transforming the dataset into complex-valued form. We tackled this using iHSV, a novel color model developed by Prof. Koteswar himself, which allowed us to encode image data into the complex domain by capturing amplitude and phase information. Implementing iHSV was a meticulous process, as we had to ensure no critical details were lost during transformation.

The next challenge was designing and training the complex-valued network. Traditional architectures weren't built for complex arithmetic, so we had to reimagine layers, activations, and optimization strategies. The early stages were frustrating the network refused to converge, and debugging felt like solving a maze in the dark. But after countless iterations and late nights, we succeeded. The final model not only performed well but outperformed its real-valued counterpart in semantic segmentation tasks. That moment of breakthrough was exhilarating. It wasn't just about the results-it was the process of pushing boundaries, experimenting, and creating something new.

2. Industry Exposure: U.S.-Based Startups: After my research experience, I wanted to understand how ML worked in corporate settings, so I pursued two remote internships with U.S.-based startups. The first was with Spotonix, which I secured through an alumni referral. At Spotonix, I worked on optimizing business intelligence models using DSPy, a novel framework. My task was to create a more efficient solution than their existing OpenAI API-based system, and I successfully achieved this by leveraging DSPy's capabilities. This experience taught me how to adapt ML models to meet specific business needs and optimize performance under real-world constraints.

The second internship was with Smarttrak Al, which I found through IIT Kanpur's internal mailing network. Here, I worked on developing Physics-Informed Neural Networks (PINNs) for fault prediction in solar inverters. I modeled the inverter systems in Python and integrated them into PINNs to enhance their diagnostic capabilities. This project was a perfect blend of theoretical knowledge and practical application, and I enjoyed the challenge of creating solutions that had tangible real-world impact.

Closing Thoughts

Looking back, my journey has been a constant exploration of how mathematics and technology can come together to solve meaningful problems. From time series predictions for Indian Railways to cutting-edge research on complex-valued networks and industry projects with U.S.-based startups, every experience has pushed me to think deeper, learn more, and tackle challenges head-on. Machine learning has become more than just a field of interest, it's a space where I can combine my love for problem-solving with a drive to make a difference. As I continue this journey, I'm excited to see where this blend of curiosity and determination will take me next.