AI Agent Engineer -- RAG, LangChain, Python (LATAM, Fractional)

ABOUT US We are a boutique AI strategy and implementation firm serving mid-market commercial real estate operators. We design and deploy agentic AI systems and workflow automation that streamline core CRE workflows: investor relations, capital raising, asset management, underwriting, and reporting. Our clients are institutional real estate firms managing portfolios of 15 to 50+ assets across multifamily, industrial, office, and hospitality. Our proprietary platform serves as the central intelligence layer for our client deployments. You will build standalone agents and AI systems that interact with this platform via APIs, MCP, webhooks, and other integration patterns. THE ROLE We are hiring an AI Agent Engineer to design, build, and deploy production-grade AI agents and LLM-powered systems for our commercial real estate clients. The work spans document understanding across messy file structures, RAG pipelines over decades of underwriting history, deal screening agents, automated report generation, and investor relations automation. This is a behind-the-scenes builder role. You will not be client-facing. You report directly to the founder and work off written specs and briefs. Hours are flexible and scale with project load, typically 10 to 20 hours per week on retainer. WHAT YOU WILL DO - Design and build standalone AI agents that integrate with our proprietary platform and client systems via APIs, MCP, and webhooks - Build and optimize RAG pipelines: document ingestion, chunking strategies, embedding, retrieval, and response generation across heterogeneous CRE document sets (rent rolls, T-12s, offering memoranda, quarterly reports, underwriting models) - Write production-grade prompts and instruction sets that perform reliably and consistently - Build document parsing and extraction pipelines for PDFs, spreadsheets, and unstructured file systems - Develop agent orchestration workflows: multi-step reasoning, tool use, human-in-the-loop patterns - Deploy and maintain containerized agent services in production - Document architecture, prompts, and system behavior so it can be handed off and maintained REQUIRED - Python as primary language - Production experience with LLM APIs (Anthropic Claude and/or OpenAI) - LangChain or LangGraph for building structured LLM applications - Vector database experience: pgvector, Pinecone, Weaviate, Chroma, or similar - Agent orchestration frameworks: CrewAI, AutoGen, Pipecat, or similar - RAG pipeline design and implementation (not just toy examples; real document sets with messy, inconsistent formats) - Prompt engineering with a focus on reliability, consistency, and structured outputs in production - Docker for containerization and deployment - REST APIs, webhooks, MCP for integrating agents with external systems - Git for version control and collaboration - Strong written and spoken English; you will work off written specs, join calls, and document what you ship NICE TO HAVE - AWS experience (our platform infrastructure runs on AWS) - n8n or similar workflow platforms (useful for triggering agents from automation pipelines) - Document parsing libraries and tools: unstructured.io, LlamaParse, PyMuPDF, Camelot - Experience with financial document formats: rent rolls, T-12 operating statements, investor reports, underwriting models - Commercial real estate or financial services exposure - Familiarity with MCP (Model Context Protocol) server development - Experience building human-in-the-loop approval workflows for AI outputs ENGAGEMENT DETAILS - 1099 contractor, fully remote - Latin America preferred (US timezone overlap is important for collaboration) - Fractional retainer, typically 10 to 20 hours per week; scales with project load - $35-55/hr depending on experience HOW TO APPLY We read every application, but we filter hard. To be considered, your application must include all of the following: 1. Start your message with the word "Agent." This tells us you read the full posting. 2. Record a Loom (5 minutes max) walking through an AI agent or RAG system you built that ran in production. Show us the architecture, explain your chunking/retrieval strategy, walk through how you handled edge cases in real documents, and tell us what failed and how you iterated. Screen share your code or a diagram; do not just talk over slides. 3. Link to a GitHub repo or portfolio with relevant work. We want to see your Python, your prompt structures, and how you organize agent code. Private repos are fine; just grant access and include the link. 4. Your hourly rate and typical weekly availability. Applications without the Loom will not be reviewed. We are evaluating spoken English, technical depth, and whether you can explain your own architecture decisions clearly.