Staff Machine Learning Engineer, Okta

• 8+ years of software development experience, with strong programming expertise in Python (and familiarity with Go or Typescript a plus)
• Hands-on experience with applied machine learning, from feature engineering to training and fine-tuning models
• Hands-on experience with modern Generative AI platforms (AWS Bedrock, OpenAI, Anthropic, etc.)
• Deep understanding of retrieval-augmented generation (RAG), embeddings, and knowledge-base workflows
• Hands-on experience with LiteLLM, LangGraph, LangChain, LlamaIndex, MCP, or other related AI agent frameworks
• Familiarity with ML frameworks (FastAPI, PyTorch, TensorFlow, Spark ML) and workflow orchestration tools (Airflow, etc.)
• Experience defining evaluation metrics, pipelines, and feedback loops for ML/GenAI systems
• Proven ability to collaborate with product and engineering teams to drive greenfield initiatives forward, navigate unknowns, and iterate quickly and frequently
• Experience building tools or infrastructure for AI/ML applications, with a deep understanding of the developer lifecycle in an AI-native world
• Education:Bachelor’s or Master's degree in Computer Science or related field

Desirable

• In this role, you will join the Agent Access Policies sub-team within Okta Secures AI as a Staff Machine Learning (ML) Engineer to advance Okta's authorization capabilities
• By replacing static, rule-based systems with dynamic AI security mechanisms, you will deliver real-time threat inspection, agent intent evaluation, and behavioral analysis—ensuring autonomous AI agents operate safely within specified bounds
• Implement intent-based enforcement to verify agent runtime requests match their intended purpose, requiring key capabilities
• Apply LLM reasoning and prompt parsing to interpret prompts, tool payloads, and intent in real time
• Integrate low-latency inference or semantic evaluation engines directly into the API gateway request path
• Use embeddings, vector search, or zero-shot classification to score alignment between agent intent and executed actions
• Design confidence-scored decision engines that feed semantic verification results into policy frameworks (e.g., Cedar)
• Establish evaluation benchmarks, prompt injection defenses, and guardrails to prevent bypasses or false positives
• Architect scalable ML and Generative AI systems integrating retrieval, inference, and evaluation pipelines
• Optimize prompting, context retrieval, and RAG workflows for accuracy, safety, and efficiency in Claude-based systems
• Build automated evaluation pipelines to measure model quality, correctness, groundedness, and safety in production
• Implement schema validation, structured output enforcement, and guardrails to ensure reliable, compliant AI outputs