[Responsibilities] Design and implement autonomous agent architectures that handle multi-step reasoning, tool use, memory, and planning in production security workflows Build reliable agent loops with proper error handling, retry logic, guardrails, and human-in-the-loop approval gates for high-risk actions Develop dynamic tool-calling pipelines where agents select, configure, and orchestrate external security tools based on contextual analysis Engineer prompt chains and agent reasoning strategies across multiple LLM providers (OpenAI, Google Vertex AI, Anthropic Claude) with model-agnostic abstractions Build and maintain the context layer that gives agents awareness of users, assets, past incidents, typical behavior, and environmental state Design evaluation frameworks to measure agent reliability, accuracy, and safety — especially for high-stakes actions like exploit execution or incident response recommendations Collaborate with security engineers to translate offensive and defensive domain expertise into agent behavior, tool profiles, and decision logic Optimize for latency, cost, and token efficiency in production agent workloads Support on-premise deployments using self-hosted open-source models (DeepSeek, Llama) for air-gapped enterprise customers [Requirements] Must Have 4+ years of software engineering experience with strong proficiency in Python 1+ year of hands-on experience building LLM-powered agent systems — not chatbots, but agents that reason, plan, use tools, and take multi-step actions Deep working knowledge of at least one agentic framework: LangGraph, LangChain, wAI, AutoGen, or equivalent Experience with tool-calling / function-calling patterns, including dynamic tool selection and chaining Solid understanding of prompt engineering for complex reasoning tasks — chain-of-thought, ReAct, plan-and-execute patterns Experience integrating with multiple LLM providers (OpenAI, Anthropic, Google) and managing model-agnostic abstractions Ability to build robust, production-grade systems — not just prototypes. You think about failure modes, retries, timeouts, guardrails, and observability Strong fundamentals in distributed systems, async programming, and API design Strong Plus Background in cybersecurity — offensive (pentesting, vulnerability assessment, red teaming) or defensive (SOC operations, SIEM, detection engineering, incident response) Familiarity with MITRE ATT&CK framework, Sigma rules, or detection-as-code practices Experience deploying and fine-tuning open-source LLMs (Llama, DeepSeek, Mistral) for on-premise or air-gapped environments Knowledge of graph databases (Neo4j) for modeling attack paths or entity relationships Experience building multi-agent systems where multiple specialized agents collaborate, delegate, and share context Familiarity with RAG pipelines, vector databases, and knowledge retrieval systems Mindset You're excited about building AI systems that take real action in the world, not just generate text You have a strong security intuition — you think about what can go wrong, what an adversary would do, and where guardrails are needed You're comfortable with ambiguity and can translate vague domain requirements into concrete agent behaviors You care about reliability over cleverness — a working agent that handles edge cases beats a flashy demo