ML Research Scientist

Theorem; San Francisco (CA); On-site; USD 300,000 - 500,000; Full time

Job summary

Theorem is building software correctness feedback that scales with the complexity and volume of code, so software is provably safe and reliable. We’re building for a world where AI raises the ceiling on software engineering.

This role focuses on post-training methods that help models perform program analysis and formal reasoning, and on interfaces that help users find bugs, generate design documentation, and implement non-functional requirements using compiler‑level reasoning.

Job description

We’re building products to make the future of software correct, understandable, and secure. Right now, AI has raised the floor on software engineering. We’re building for the world where it raises the ceiling.

What We’re Building

Our approach is to write software correctness feedback that scales with complexity and volume of code, so software is provably safe and reliable.

What This Looks Like In Practice

• Post-training that make models as good at program analysis and formal reasoning as they are at writing Python.
• Interfaces for navigating complex implementations of simple specifications to help users find bugs, generate design documentation, and implement non-functional requirements using compiler-level reasoning methods.

You may be a good fit if you

• Pick projects that are a few orders of magnitude bigger, faster, or harder than SOTA
• Have the gumption to try out something before everyone starts talking about it
• Love to be helpful to the people around you, whether it is pair-debugging or teaching someone something new or just picking up the slack on the mundane tasks that are blocking progress
• Dig deep to get satisfying explanations for how and why things work
• Are practiced in solving last-mile problems

Compensation

$300,000-500,000 with generous equity