Real estate underwriting agent

What we built

We built a real estate underwriting agent that handled the core steps of an end-to-end underwriting workflow instead of just solving one narrow part of the process. The system was designed to speed up analysis while still surfacing the places where human review should stay involved.

The workflow moved from operating memo to model inputs, surfaced flags for human review, and then generated a clean synopsis writeup so the team could move from raw deal materials to a more decision-ready package with less manual coordination.

Workflow coverage

• Operating memo ingestion. Parsed source materials and extracted the key details needed to start underwriting.
• Model input preparation. Structured the relevant assumptions and values so they could flow into the underwriting model.
• Human review flags. Marked issues, edge cases, and assumptions that warranted manual attention before finalizing output.
• Synopsis writeup. Produced a concise summary of the deal and the underwriting conclusions for faster review.

Data and comp gathering

• Scraped market and submarket data to support the underwriting workflow.
• Collected rental comps as part of the analysis process.
• Allowed rental comps to be selected both by AI and by human input, so the team could guide the system when context mattered.

Analysis layer

• Ran scenario analysis across key assumptions that materially affected returns.
• Performed sensitivity analysis on major variables so the team could understand downside, upside, and risk concentration.
• Created a workflow that balanced speed with reviewability rather than hiding assumptions inside a black box.

The result was an underwriting process that gave the team a much stronger first pass on each opportunity while keeping critical judgment calls visible, reviewable, and easy to refine.