AI Engineer- Responsible AI

Descrição da Vaga

Requisitos

• 2+ years of industry experience in applied ML/AI research or ML engineering
• Track record of publications in AI Safety, NLP robustness, or adversarial ML (ACL, NeurIPS, ICML, EMNLP, IEEE S&P, etc.) or equivalent applied research impact
• Strong Python and PyTorch/JAX skills with experience deploying ML models to production
• Demonstrated experience in at least one of: LLM jailbreak attacks/defense, agentic AI safety, adversarial ML, or human-AI interaction vulnerabilities
• Experience with containerization (Docker, Kubernetes) and cloud platforms (AWS, GCP, or Azure)
• Proven ability to take research from concept to code to production deployment with rigorous testing and monitoring
• Safety: Red-teaming frameworks, LLM benchmarking (SODE, ART, HarmBench), human behavior simulation
• Infrastructure: Kubernetes, Docker, Terraform, AWS/GCP, Ray, Spark
• Production systems that measurably improve safety KPIs: adversarial robustness, over-defensiveness rates, and incident response latency
• Publishable research outcomes (with company approval) demonstrating novel contributions to AI safety
• Infrastructure that scales reliably and enables the broader team to iterate quickly on safety research
• Mentorship: Collaborate with Principal Architects and senior researchers in AI safety and adversarial ML
• Velocity + Rigor: Balance high-quality research with mission-critical product focus

Responsabilidades

• Build the Future of Safe and Responsible AI
• Advance AI Safety: Design, implement, and evaluate attack and defense strategies for LLM jailbreaks (prompt injection, obfuscation, narrative red teaming) and deploy them as production-grade services
• Build Scalable Safety Infrastructure: Architect and deploy distributed safety evaluation pipelines handling millions of requests, with real-time monitoring, alerting, and incident response capabilities
• Large-Scale Data Engineering: Design ETL pipelines for processing terabytes of safety-related data (attack patterns, behavioral logs, model outputs); build data lakes and feature stores for safety ML systems
• Evaluate AI Behavior: Analyze and simulate human-AI interaction patterns at scale to uncover behavioral vulnerabilities, social engineering risks, and over-defensive vs. permissive response tradeoffs
• Agentic AI Security: Build production workflows for multi-agent safety (agent self-checks, regulatory compliance, defense chains) spanning perception, reasoning, and action
• MLOps & Model Deployment: Deploy safety models to production using containerized microservices, implement CI/CD pipelines for model updates, and manage model versioning and A/B testing infrastructure
• Benchmark & Harden LLMs: Create reproducible, automated evaluation protocols for safety, over-defensiveness, and adversarial resilience across diverse models with continuous integration
• Production Red-Teaming Platform: Build and operate an automated red-teaming infrastructure that continuously probes advanced LLMs (GPT-4o, GPT-5, LLaMA, Mistral, Gemma) at scale, with dashboards and alerting
• Real-Time Defense Systems: Implement context-aware, multi-turn attack detection and guardrail mechanisms as low-latency services handling 10K+ requests per second
• Agent Self-Regulation at Scale: Develop agentic architectures for autonomous self-check and self-correct with distributed orchestration and fault tolerance
• Safety Data Platform: Design and build data infrastructure for collecting, storing, and analyzing petabyte-scale safety telemetry with streaming analytics
• MLOps: Triton Inference Server, Weights & Biases, MLflow, Airflow, ArgoCD
• Well-documented, tested, and maintainable code with comprehensive CI/CD and monitoring

Benefícios