AI Engineer

About Crexi

Commercial real estate is one of the largest asset classes on earth and one of the last to go digital. Deals still get done over email threads, PDFs, and relationships nobody can search for. Crexi is the exchange betting that doesn't have to be true anymore: we're building the platform where an entire industry finds, evaluates, and closes deals, and we're betting hard on AI to get there faster than anyone thought this industry could move. There's no playbook for turning a trillion-dollar, relationship-driven industry into a modern marketplace which means the people building it get to write it.

About the Role:

The AI Engineer builds the agentic AI systems that power Crexi's platform, including orchestration, retrieval, and action layers grounded in Crexi's proprietary commercial real estate data. The role builds multi-step agentic workflows that understand user intent, orchestrate the right capabilities, and complete real work, not just answer questions, using tools like LangGraph and AWS Bedrock AgentCore. It exists to extend Crexi's AI-powered research, document generation, and zoning intelligence into unified, trustworthy agentic experiences for brokers, appraisers, lenders, and investors. 

What You'll Do:

A typical day may include: 

• Design and implement multi-step agentic workflows, including routing, planning, tool use, and state management, using frameworks like LangGraph/LangChain to move from user intent to real, completed actions. 
• Build evaluation frameworks and production monitoring (offline evals, human-in-the-loop annotation) to measure quality, reliability, and business outcomes, not just model metrics. 
• Build context and retrieval pipelines over Crexi's proprietary CRE data, handling the compound, qualitative queries that structured filters can't answer. 
• Work primarily with Anthropic frontier models on AWS Bedrock, using Bedrock AgentCore for agent runtime, identity, and memory, and helps evaluate open-weight alternatives on cost, latency, and capability. 
• Implement guardrails, structured outputs, and graceful ambiguity handling so agentic systems ask for clarification rather than acting on unclear intent, and establishes trace-level observability across LLM calls and tool invocations. 
• Leverage agentic coding tools (e.g., Claude Code, Codex) for scaffolding, refactoring, test generation, and debugging, while critically reviewing AI-generated output for correctness, security, and performance. 
• Rapidly prototype against real user feedback, partnering with Product and Design, and helps convert successful prototypes into scalable, reusable capabilities. 
• Ensure AI outputs are explainable, source-linked, and trustworthy enough to inform real business decisions.