I grew up in Bangalore, India, and I’ve seen how tech and capital can be hugely transformative for individuals and societies. I try to have opinions on AI, startups, politics, markets, and tech. (Also TFT sets and esports.)
I spent last summer working on Squire (with Harish). Squire was our vision for AI assistants at work. Squire began as a set of fine-tuning experiments (and we tried out some distillation) to make a general-purpose assistant that lived on my computer and took care of work that annoyed me (mostly routine admin). We were targeting small models that we could fine-tune for specific tasks and then swap them for the task at hand. We realized that localized inferencing wasn’t worth the limitations it posed for our use cases. The excitement was more technical awe at how far it had come along.
We spent time building Squire to be the AI for all admin. You text, call, or email Squire and it takes care of scheduling, mailing, researching, coordinating, and filling out forms. We wanted Squire to be the bridge between the present and the truly autonomous systems of the future. More here.
Squire is no longer active, mostly because the pivots we found ourselves facing didn’t excite us, and we decided we’d rather work on exciting tech right now than force it. I learnt a lot from Squire and have a better sense of what I would do differently next time around.
I’ve spent time on distillation experiments, fine-tuning LLMs (how Squire started), evals for agents, and exploring the different libraries and protocols related to AI systems. I got into computer science in middle school while flashing custom Android distributions onto my S4 mini and competing in the First Lego League.
I also have some experience with business development and marketing, academic research (in economics and the social sciences), content-based SEO, large-scale data analysis, consulting, and software engineering.