Job Details

Research Engineer – Formal Methods & Reasoning Systems (AI Safety Focus)

Acceler8 Talent is working with an AI research lab to explore new directions in AI alignment and interpretability, with an emphasis on moving beyond purely empirical techniques toward more rigorous, structured approaches. They are interested in incorporating ideas from fields such as programming languages, formal verification, and program analysis to better understand and reason about model behavior.

This position is well-suited for individuals who may not have a traditional machine learning background but bring strong expertise in formal systems, abstraction, and rigorous reasoning, and are interested in applying those skills to AI safety challenges.

Key Responsibilities

• Investigate how concepts from formal verification and program analysis can be applied to understanding the internal behavior of machine learning models
• Develop experimental tools that integrate structured reasoning techniques with interpretability methods
• Work closely with research teams to design new abstractions for analyzing and constraining model behavior
• Prototype approaches that move toward verification-inspired safety assurances for AI systems
• Contribute to shaping emerging research directions within the organization

Example Areas of Work

• Adapting compiler-inspired abstractions to represent and analyze internal model computations
• Creating structured methodologies for reasoning about activation-level behavior and model circuits
• Building tooling that combines interpretability techniques with constraints influenced by verification principles
• Exploring program analysis techniques to identify potentially unsafe reasoning patterns within models

Qualifications

• Strong background in areas such as programming languages, compilers, systems research, formal verification, or related fields
• Experience in theorem proving, security research, operating systems, or similar domains is valuable
• Demonstrated ability to think abstractly and build novel research systems or tooling from the ground up
• PhD is preferred