Thesis is an applied AI lab building an intelligent platform that collaborates with humans to invent, train, and refine machine learning models.

We focus first and foremost on machine learning as it exists in the real world: iterative, experimental, and performance-driven. From insurance to quantitative finance to high-impact production ML systems, progress today depends on fragmented tools and individual intuition. Thesis turns that process into a systematic, AI-native loop that compounds over time.

Our mission is to democratize frontier machine learning by giving researchers and engineers a shared environment where human judgment and intelligent systems work together to deliver real, measurable improvements.

Machine Learning R&D

Machine learning advances through iteration: choosing data, designing architectures, running experiments, interpreting results, and deciding what to try next. There is no closed-form theory for this process. Progress comes from making good decisions under uncertainty.

Thesis systematizes this loop.

We are building systems that collaborate with humans end-to-end to invent models, train them at scale, debug failures, and refine performance. The platform observes experiments, decisions, and outcomes, learning which actions lead to real gains and which do not. Over time, this turns hard-won engineering intuition into reusable signal.

Benchmarks such as OpenAI's MLE-Bench and METR's RE-Bench point toward the next phase of ML: agents that don't just write code, but conduct machine learning research. Thesis is the environment where those agents operate alongside humans.

The Vision

In the limit, Thesis becomes the default environment for applied machine learning.

Any team should be able to run thousands or millions of experiments in silico, guided by intelligent agents that reason over data, models, and prior results. Machine learning progress should compound across projects, organizations, and industries.

As ML becomes the substrate for decision-making in finance, insurance, and every data-driven system, the way we build and improve models will determine who moves fastest. Our goal is to define the platform that makes that progress systematic.

Collaboration

We believe the future of machine learning is collaborative, transparent, and accelerated by AI.

Our work is shared through technical writing, open-source software, and partnerships with applied ML teams and researchers pushing models into production.

If you are building machine learning systems and care about improving them rigorously, we'd love to collaborate.

- The Thesis Team, 2025