Role overview
At Fundamental, we believe that the world's most critical decisions, from fraud detection to supply chain logistics, are made on tables, not text. We are building the world’s first Universal Predictor, a Foundation Model designed to outperform traditional approaches on high-value enterprise use cases.
We aim to prove value in the field by moving quickly from research breakthroughs into real customer deployments. We are developing a technology designed to operate reliably within the complex constraints of enterprise infrastructure, ensuring security and measurable business outcomes.
We are a rigorous, collaborative team bridging the gap between research and reality. Our workforce thrives on solving hard integration challenges and identifying the intersection of data feasibility and business impact. We are pragmatic, ambitious, and focused on value creation.
Join us to shape the future of tabular AI and turn bespoke customer solutions into a scalable platform.
What you'll work on
- Take part in development and optimization of a large neural network-based tabular model implemented in Python
- Profile training and inference pipelines to identify performance bottlenecks
- Rewrite critical components in C++ (via PyBind11 or custom extensions) where Python limits us
- Improve memory efficiency, latency, and throughput across model pipelines
- Ensure correctness, numerical stability, and reproducibility as the model evolves
- Collaborate with ML researchers on productionizing new capabilities
- Maintain clean abstractions, comprehensive tests, and clear documentation
- Shape architectural decisions for our ML systems handling tabular data
What we're looking for
- Experience with tabular ML approaches (transformers, tree/NN hybrids, learned embeddings)
- Familiarity with PyTorch internals or writing custom ops
- Experience optimizing training loops, data pipelines, or inference engines
- Background in numerical computing or systems programming
- Exposure to large-scale ML infrastructure (distributed training, batching, caching)