Senior AI Engineer

Crexi is reimagining commercial real estate with an AI-powered platform built to deliver smarter, more efficient solutions at every stage of the deal lifecycle. From real-time data and market insights generated by Crexi Intelligence, to targeted property marketing and seamless deal management through Crexi PRO, and a transparent, time-bound bidding experience with Crexi Auction— Crexi enables users to evaluate opportunities, maximize exposure, and close with speed and confidence.

To date, Crexi has facilitated over $1 trillion in transactions, 8.6 billion square feet leased, and supports a growing community of more than 2 million monthly active users.

Crexi’s mission is to catalyze the next generation of commercial real estate through three core pillars:

Access, Innovation, and Connection. Crexi’s platform democratizes CRE by providing unprecedented access to market insights and opportunities, accelerates CRE dealmaking with purpose-built technology that enhances speed and transparency; and empowers CRE professionals with a centralized platform designed for real-time collaboration and success.

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 Lang Graph and AWS Bedrock Agent Core . 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 Lang Graph/Lang Chain 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 Agent Core 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.

Qualifications:

• 5+ years of professional software development experience in production environments with meaningful scope/ownership, including 2+ years hands-on building LLM-powered applications in production.
• Real experience with agentic architectures (multi-step pipelines, tool/function calling, state machines, memory), ideally with Lang Graph or similar orchestration frameworks.
• Strong retrieval design skills: RAG systems, embeddings, vector search, and the judgment to know when retrieval is the problem and when the data is.
• Strong Python proficiency and real-world experience building backend services and APIs in production.
• Experience with cloud infrastructure (AWS preferred) and modern collaborative workflows (Git Hub/Git Lab, code reviews, CI/CD).
• Demonstrated use of agentic coding tools (e.g., Claude Code, Codex) in a professional workflow, without compromising quality.
• Comfortable operating in ambiguity and fast iteration cycles;…