As my undergraduate chapter closes, I’m reflecting on the unpredictable path of discovery that shaped my interdisciplinary AI journey. My journey was anything but linear; it was a winding road of pivotal moments that often felt like detours at the time. But stepping into my new role as a Lead ML Engineer at Neurogati, an IIT Madras-incubated startup, feels like more than a transition—it’s a convergence. It’s the realignment of a purpose I’ve been slowly discovering all along. I am ready to embrace this next chapter, channeling every lesson from my unique path into building human-brain-inspired AI and contributing to the pursuit of Artificial General Intelligence (AGI).
The Genesis – Finding a Direction in Tech
My real exploration of computer science began in my second year when I laid out the core domains for exploration: Cyber Security, AI, and 3D Development. While I was never a pure hardware person, I loved making hardware work through software, which led me to build a home server for my lossless music library, movie collection, and personal cloud storage (a story for another time). I dove into everything—taking ethical hacking courses, starting simple ML projects, and even completing the mandatory “Blender Guru Donut Tutorial“. As I got deeper, I realized that AI was the common thread that could weave all my interests together. At the same time, 3D modeling and environment design felt like a natural, digital extension of my background in traditional art. It was just another medium for creative expression. The timing was perfect, as the emergence of early Large Language Models like GPT-2 provided the final spark that fully ignited my passion for AI.
Where Worlds Collide – The Interdisciplinary Superpower
My interest in photography, which began at 14 with my father’s Nikon D60, gave me an unexpected edge. What I learned about composition and lighting from devouring issues of “Better Photography” magazine directly translated to setting up virtual cameras in Blender. This synergy between the real and virtual worlds nudged me from Blender toward the more cinematic Unreal Engine.
A similar transition happened with my music. The theoretical knowledge from my Trinity Grade 8 exams in Piano and Theory of Music found its practical voice in the world of music production, orchestration, and synthesizers. This fascination with synthesizers and analog effects led me down the rabbit hole of audiophile communities, where I learned to appreciate the subtle randomness and warmth of analog systems—that imperfect, human feel that digital precision often misses.
It was through these experiences that I had my “Aha!” moment: these fields aren’t separate at all. They are interwoven with overlapping patterns and principles. This realization solidified the vision for my work in AI: to combine AI’s precision with human intuition. This realization was the true foundation of my interdisciplinary AI journey.
Laying the Groundwork: First Steps in Applied AI
That “Aha!” moment was the catalyst. It was time to transition from merely enjoying the patterns to actually constructing with them. My journey into real-world AI started with an internship at Bunjee Tech, which was my first actual experience with machine learning at an industrial scale. I was thrown into the deep end, working with huge 5TB datasets and deploying transformer models with k-Nearest Neighbors (k-NN). It was a crash course in managing data at a level I never thought I’d reach and it cemented my core knowledge.
From there, I embarked on my first serious independent project: Medical Image Classification. This is where the abstract deep learning concepts really came alive. I spent hours wrestling with hyper-parameter tuning, carefully adjusting learning rates, batch sizes, and optimizers. It was an exercise in patience and accuracy, learning the art of tuning models to eke out every last bit of performance from the GPUs.
Tackling the Modalities: Text, Audio, and Vision
Text
The next challenge was the most important one of my undergraduate life: creating a chatbot by training a Large Language Model for the Indian Judicial System. The challenges were immediate and monumental. The most obvious problem was the restrictive context window of models like BERT, which was simply too small for long legal documents. This compelled me to roll up my sleeves and essentially reverse-engineer the model’s architecture to accommodate a vastly larger context than the default 512 tokens. Just as I was getting traction, the entire scene changed with the advent of mighty open-source LLMs like Mistral and Llama. This was a game-changer. I had to adapt, creating thoroughly planned data preparation, training, and inference pipelines to deal with their giant parameter counts.
Audio
Having mastered the text modality, my interests logically led me to audio. Inspired by Google’s “MusicLM” paper and fueled by a love for music production, I set out to develop a text-to-music generation model. The journey was far from straightforward; the paper didn’t specify the precise architecture. This required me to construct the fundamental building blocks from scratch—not one, but three separate, hand-coded transformer models, each performing a very specialized function within the generation pipeline. During this project, I also learned about the fascinating field of neural codecs, which can losslessly encode and reconstruct audio, a technique I incorporated to maintain the quality of the generated music.
Image
With text and audio explored, only the image modality was left. My last significant project was an interactive image captioning system. This experience taught me how to handle and process large-scale image data effectively to train Vision Language Models (VLMs). I worked with some of the state-of-the-art models like CLIP, BLIP, and most notably, LlaVa—a robust open-source model that was directly competing with behemoths like OpenAI’s GPT-4 Vision.
Synthesizing an Interdisciplinary AI Journey: The Road to Neurogati
With an assortment of varied, multi-disciplinary projects in my portfolio, I needed a way to tell my story cohesively. I created my personal website not as a resume, but as a story—a space to tastefully share how these seemingly disparate interests came together in one purpose.
That synthesis found its truest form during my internship at the Computational Neuroscience Lab in IIT Madras. This was the setting I had been looking for—an interdisciplinary setting where computer science and neuroscience weren’t just neighboring fields, but heavily interconnected. It was here that everything fell into place. My task was to create a hybrid model combining a Graph Neural Network (GNN) and a Large Language Model. The project was the ideal culmination of my interdisciplinary AI journey, asking me to tap into all my experience with different data modalities and stitch them together into one cohesive, complicated system.
Something rather unexpected also occurred during this period. My creative and technical endeavors diverged into two separate but parallel tracks. In one track, I started writing this blog to communicate and share my experience. At the same time, as an entirely independent task, I embarked on the challenge of designing a mobile app from scratch. This project necessitated the development of a completely new set of skills, learning front-end development in React Native, designing a strong backend with Node.js, and hosting the whole system on the AWS cloud infrastructure. It was a time of frenzied, multidimensional learning, affirming the belief that development is exceptionally accelerated by undertaking varied and dissimilar challenges.
The Next Frontier
And so, I stand here today, at the end of an incredible interdisciplinary AI journey and on the threshold of this new chapter as a Lead ML Engineer at Neurogati. And as I look back, the journey was anything but linear. It was a twisted path of discovery, fueled by an insatiable curiosity that never chose to remain in one lane.
I’m entering this new position not only with a technical skill set, but with a distinctive narrative born of curiosity, chaos, and convergence. My artistic and musical background is not an appendix; it’s the basis of my intuition. The many late nights spent tuning models, writing transformers from the ground up, and reverse-engineering architectures have developed a strong technical discipline.
My vision is sharper than ever. I am poised to direct this special combination of creative insight and technical skill into the next great challenge: constructing human brain-inspired AI. The ambition is not merely to develop smarter algorithms, but to make a significant leap forward in the great quest for Artificial General Intelligence. The next frontier awaits…
Leave a Reply