“Tabular foundation model”: I am building & innovating in this space. This is an important video for all the people in my circle to know and understand. I have been writing about why the only defensible moat is “accrued knowledge” & no “engineered harness” comes close to it. Building a specialized foundation model is one of the most important and challenging task. I am proud of myself and my team to continue to stay updated around these changing times. I moved to SF & its one of the most important part of making this idea successful. Before pushing this post, I was waiting for some work to be done on our end and we have started to now hill-climb. There are and will be challenges but we are pushing through those walls like a breeze now because of the intellectual depth we have as a team. I want to highlight few more important things here related to the research part. I want to take pride in saying that we have been now matching the mindset of the frontier research in this field. The approaches generally that we come out are mostly end up being available in arxiv. The beautiful part is that these all research are mostly <3-4 months old. It means that we are almost on the frontier of this game. And since we thought of it originally completely out of our personal Pain, we have original approaches to solve it and we are not dependent on these research papers but seeing these research papers validate our approaches that we want to take. A bit about how this problem came into existance: this problem was originated very organically when I was teaching myself Quant in the last 6-7 months PLUS when we did an internal exercise of whether these AIs can handle the time-series data or not. And we found out that it can’t. It understands a bit of volatility but it just don’t understand the structured data in superb manner the way it does with the unstructured data. And when we went a layer deep then we found out that transformer architecture doesn’t have the concept of time in it. It has attention but with no time concept. Overall, The actual moat which essentially is going to catapult us in the league of those companies who are innovating in this space is our quality of thoughts & approaches which is helping us now test those thesis at superfast speed. Its super challenging & exciting. We are gunning with everything. And hopefully soon, you will see us producing benchmarks with variable infrastructure and with reproducible results. See you on the next summit!