AI Safety & Reasoning Engineer (Reasoning Transparency)

Sourced from scoutify.com

• Operationalize transparency by translating mechanistic interpretability concepts into scalable inference-time pipelines
• Build infrastructure to extract, map, and interpret model activations and latent representations
• Design red-teaming frameworks and automated behavioral auditing stress tests for misalignment, sycophancy, and deception
• Partner with ML engineering to integrate transparency probes while managing latency, memory, reliability, and cost
• Track AI safety research and prototype, evaluate, and deploy promising transparency methods
• Define evaluation methods and communicate evidence, limitations, and findings to cross-functional partners

• Bachelors + 7 years, Masters + 4 years, PhD + 1 year of related experience, or equivalent
• Hands-on experience deploying or operating ML models in production using Python
• Ability to conduct rigorous empirical work and communicate technical results
• Practical understanding of transformer-based language models, inference-time computation, and alignment approaches such as RLHF or RLAIF
• Proficiency building evaluation frameworks with adversarial prompt engineering

• High proficiency in Python frameworks for distributed training or high-throughput inference
• Experience probing or analyzing model activations, circuits, and representations
• Experience with Kubernetes, CI/CD tools, and model tracking tools
• Information security principles or structured threat modeling applied to AI systems
• Familiarity with methods for eliciting model knowledge and evaluating output/internal-signal divergence