Software Engineering Intern

The average real estate agent closes about one deal a month. Not because they can't sell — because they're buried in disclosures, coordination, chasing signatures, and follow-up that nobody pays them for. We're building the AI agent layer that absorbs all of it, so an agent can run 10 deals a month without a transaction coordinator and without cutting corners.

That means shipping real AI agent systems into production, in a regulated industry, where being confidently wrong has consequences. Not a chatbot demo.

• —Ship features on both sides of the marketplace in our React app — the seller and buyer experience, document flows, agent admin panel, and internal dashboards.
• —Build API endpoints in FastAPI and design schema and queries against Postgres, using SQLAlchemy and Supabase.
• —Build AI agents end to end: prompt design, tool calling, structured outputs, and grading agent behavior against real call and chat transcripts.
• —Use AI tools to write code.
• —Make AI workflows trustworthy: catch hallucinations, tighten branching logic, and force non-deterministic systems into predictable behavior.
• —Deploy and troubleshoot on GCP using Cloud Compute, Cloud Run, Cloud Build, and Docker.

• —Solid grasp of TypeScript and React.
• —Comfortable writing Python.
• —You've built and launched something real — side project, hackathon, or class project all count.
• —Can model relational data and write your own SQL.
• —Know your way around REST APIs, both calling them and building them.
• —Confident with Git.
• —AI coding tools (Cursor, Claude Code, Copilot) are already part of how you work, and you've developed judgment about where they help and where they don't.

• —Hands-on time with FastAPI, SQLAlchemy, or Supabase.
• —You've deployed something to a cloud platform — GCP preferable.
• —Built with LLM APIs before: prompting, function/tool calling, RAG, or agent frameworks.
• —Written or used an MCP server.
• —Working knowledge of Docker.

• —You ship. Something rough and live beats something polished and unreleased.
• —You chase evidence when debugging instead of guessing.
• —You gather context once, then move without hand-holding.
• —Ambiguity doesn't stall you — much of what we build has no template to copy.

• —A monthly stipend.
• —Genuine ownership of features that real home sellers use, inside a regulated industry.
• —A direct line to the founding team, CEO included.
• —Real experience shipping production AI agent systems — the part of AI engineering that's still genuinely hard and doesn't come from tutorials.
• —Work you can point to in an interview, plus a reference to back it up.