Research Intern – AI Interpretability

Job Description

Research Intern: Interpretability & Reliability (Summer 2027) | CTGT

The Tone:
This is a full-time internship at CTGT, located in San Francisco. CTGT builds a deterministic governance layer for AI, enabling important global institutions to deploy AI workflows with confidence. This role is crucial because frontier AI models, while often right, are occasionally confidently wrong in ways that make them useless for high-stakes environments. This intern will help close that critical gap by owning a hard research problem to make systems provably reliable.

The TL;DR
• Role: Internship
• Type: Full-time
• Location: In-person, San Francisco, CA

• Team: Research function, sitting directly with engineers building the Policy Engine.
• Mission: Own one hard research problem inside the interpretability and reliability program from end to end.
• Tech Stack: Python, Rust, Node/TypeScript, React, PostgreSQL, vector, graph databases, Docker, Kubernetes, Terraform, PyTorch.

What You’ll Actually Do
• Implement: Stress-test methods for feature extraction and runtime intervention, ensuring they work repeatably across model families.
• Design: Create evaluations that bound error, focusing on calibration under imbalanced data and verifiable task regimes.
• Research: Read relevant literature, determine what matters, reproduce findings, and push past existing theories.
• Collaborate: Work with engineering to transform research findings into capabilities for the Policy Engine, shipping into audited, high-stakes environments.
• Present: Defend your reasoning and present progress every week in research reviews.

The Must-Haves
• Background: Pursuing a degree (Bachelor’s through PhD) in computer science, mathematics, the sciences, or a similarly unforgiving quantitative field; Student level.
• Experience: Experience writing real code for real computational systems; demonstrated fluency with PyTorch and the modern ML stack, or a track record proving quick learning ability.
• Skills: Strong mathematical foundations including linear algebra, probability, optimization, and information theory; ability to read a paper, discern relevant information, and implement findings; self-directed with the capacity to make real progress independently.
• Bonus: Drawn to interpretability, model internals, and developing systems that are provably reliable rather than just “usually fine.”