How eighteen agents converge on one source of truth
A typical Relm Pro underwriting run touches eighteen specialized agents. Each one is responsible for a single class of evidence: rent comparables, cap rate context, lease abstraction, demographic context, transit access, capex assumptions, debt service coverage, exit assumptions, sensitivity analysis. They run in parallel, then hand their findings to a synthesis layer that produces a single underwriting view.
The architecture matters because of what it makes possible at the cell level. Every number in the resulting pro forma is traceable to a public record, a market comparable, or a deep-research citation. Click any cell and you see the agent that produced it, the source it pulled from, and the reasoning chain that connected them. Nothing is a black box.
This is the inverse of how most templated platforms work. A templated pro forma assumes a market cap rate, applies a constant rent growth, and asks you to verify the assumptions are reasonable. A Relm Pro pro forma starts with the property, queries the actual market data for that submarket, and returns a number with the receipts attached. Verification work happens before you see the cell, not after.
For an analyst, the practical effect is that defensibility ships with the deliverable. Hand your pro forma to the IC and they can drill into any assumption without you re-doing the work. For a broker, your pitch deck arrives with the data already cited. For an asset manager, every quarterly review uses the same underwriting basis.
That is what we mean by "reviewable AI". The AI does the work, every edit is reviewed by a human, and every number has a source.
