Research Cofounder - Tabular Foundation Models | Knoww
We're building a Tabular Foundation Model: tables in, classification and regression out. Our goal is to make prediction accessible to banks, fintechs and other businesses without building a separate model for every task.
I'm looking for a research cofounder to build this with me, from architecture and experiments to published research and real-world deployment.
What we've built so far:
- Two signed letters of intent in finance, with more than four businesses testing the product.
- An ensemble ranking 6th in our own unofficial reruns of TabArena Lite's development tasks, not the official leaderboard.
- Early work on interpretability for banking use cases.
- A community of around 5,000 Instagram followers from documenting the research journey, helping us attract talent and banking connections.
The research:
- Learning across tabular datasets rather than training a new model for every task.
- Modeling time and relationships in structured data.
- Interpretability and recovering useful mathematical equations from model outputs.
- Rigorous evaluation of accuracy, speed and generalization.
Who I'd like to meet:
Researchers with published machine-learning work and strong experimental skills. PhDs, graduating doctoral researchers and researchers leaving frontier AI labs are particularly relevant, including people from MIT, Stanford, Harvard, Berkeley and comparable research groups.
Experience with tabular foundation models, prior-fitted networks, transformers, time-series modeling or interpretability would be especially valuable. Above all, you should want to develop original ideas, test them carefully and build a company around the work.
We've bootstrapped the work so far. This is a research-cofounder opportunity, not a conventional research vacancy.
If this resonates, message me with a paper or project you're proud of and why this problem interests you.