I work on building trustworthy AI systems at Vijil.ai, defining the trust layer for agentic AI. Just as Anthropic and OpenAI push alignment at the model layer, Vijil focuses on ensuring multi-agent systems remain reliable, verifiable, and production-ready.
Previously, I worked at Amazon Music, where I was part of the team that built the model architecture and large-scale pipelines for training and serving deep learning embedding models. We also developed a vector store to power recommendation candidate generation. These systems processed massive volumes of data, supported continuous retraining, and drove recommendation flows that served billions of user requests daily.
Earlier, I re-architected distributed AI services at Samsung's Viv Labs, advancing the infrastructure behind voice assistants. Before that, I co-founded Adya.io (acquired by Qualys) and was the founding engineer at StatX (acquired by TapClicks). I also worked at Instantis, which was later acquired by Oracle. Across these roles, my focus has been on taking ambitious ideas and turning them into production-grade systems that scale and endure.
I publish essays and primers on evaluation, agentic architectures, and trustworthy deployment — beginning with Primer on Agentic AI. Writing is how I clarify my ideas and test them against the realities of production.
I spend my time working on agentic architecture and design thinking — integrating recognition and context into the principles and practices of software engineering.
I studied at the University of Delhi, and have completed master-level courses at Stanford and Georgia Tech, along with certifications from MIT.
Outside of work, I like books, philosophy, Taiji (tai chi), and exploring the intersections of metaphysics, language, and recognition.