GenAI Solution Architect · Accenture Strategy & Consulting
Civil Engineering → Atmospheric & Oceanic Sciences (IISc) → Quant Finance → Reinforcement Learning → GenAI. On paper it looks like indecision. It isn't. I've always been chasing one thing: the joy of using math and physics to solve real problems.
I took up civil engineering out of love for mechanics, which grew into a fascination for fluid dynamics at IISc. Back in 2014–2015, statistical learning was finding widespread application everywhere and I wanted to test these ideas in industry. Quant finance gave me that playground — at DeepR Analytics I built statistical and deep learning models for high-frequency trading at the NYSE. That's where I fell in love with stochastic control and RL. From there, deep learning and reinforcement learning at minds.ai in the semiconductor space, and now leading generative AI projects across text, speech, and vision for a global energy client at Accenture Strategy & Consulting.
Beyond work, I'm associated with the Aalok Foundation, contributing to education, healthcare, and empowerment of underprivileged communities. If the learning never stops, neither should the giving back.
This journal exists because I believe we are entering the era of model interpretability — and it couldn't matter more. As language models get embedded deeper into consequential decisions, the gap between "it works" and "I understand why it works" becomes a safety problem. The intersection of interpretability and AI safety is where I want to be, and where I think the most important research of the next decade will happen.
But this page isn't just for me. It's for anyone who has built a transformer, understood the architecture, and then wondered — what's actually happening inside? What do the training curves mean? Why does one optimizer destabilize early training while another doesn't? What does stability even look like in a small language model, and how does it change when you fine-tune with RL?
These are not beginner questions. But they're also not answered in most courses or papers — you have to run experiments to find out. That's what this is. Every post here is one experiment closer to reading the black box. If you're on that path too, follow along — and reach out.