Deterministic governance for enterprise AI
CTGT builds a model-agnostic platform using mechanistic interpretability to enforce policies, prevent hallucinations, and provide audit trails for AI in high-risk enterprise environments. It converts SOPs and regulations into machine-readable rules that AI follows automatically, enabling Fortune 500 companies in finance, insurance, and media to deploy reliable generative AI. This solves the alignment and reliability bottleneck by intervening at the model's representation layer without fine-tuning or prompts.
Intern will own and ship a full project end-to-end across the stack, working closely with senior engineers on the Policy Engine and platform. Requires strong fundamentals, Python/JavaScript fluency, and comfort with Git and cloud workflows.
Designs and builds core services for the Policy Engine, owning system architecture, tradeoffs in accuracy/consistency/latency, and internal APIs. Requires 6+ years building non-trivial distributed systems with strong engineering judgment.
Develops systems for LLM interpretability and deterministic governance by working directly with model weights, activations, and architectures. Implements mechanistic interpretability techniques like activation patching and control vectors for enterprise policy enforcement in production.