Realtors need comparative market analyses to win listings, and doing one properly takes hours of pulling comps, checking photos, and writing it up. I built the AI system that generates them automatically: live MLS data in, a polished report out, with a feedback loop so agents can revise a report by simply replying to an email.
A CMA is only useful if it is both fast and right. The raw material lives in MLS systems with messy, inconsistent data; the photos matter as much as the numbers; and a first draft is rarely the final one, because the agent knows things about the property the data does not show. So the system had to pull live MLS data, judge listing photos like a human would, produce a report a realtor is proud to send, and then take corrections without anyone touching a workflow.
A layered pipeline: MLS data comes in through n8n, an LLM prompt chain and a computer-vision classifier turn it into analysis, the report goes out as HTML or PDF, and email replies feed revisions back in. Prompts live in a portal so they can be tuned without redeploying anything.
MLS lookup through the Repliers and Cornerstone APIs, address validation, find-MLS and find-comparable workflows, and handling for off-market properties. Data cleansing throughout, because MLS records lie.
Computer vision over listing photos, including stitched-image analysis, moved out of n8n into its own service so it scales independently of the workflows that call it.
The analysis lands as an HTML report with market-statistics tables, converted to PDF for delivery. The same endpoints serve the realtor Agent Assistant, so an agent can request a CMA in chat.
Every prompt in the chain is editable from a web portal, so tuning the analysis is a content change, not a deployment.
When an agent replies to a CMA email with corrections, the system parses the reply and regenerates the report. No form, no ticket, no workflow edit.
Automated tests run the pipelines against known inputs in bulk, with alerts on the comparable subworkflows, so regressions surface before a realtor sees a bad report.
What used to be hours of manual comp-pulling is now a request: the system fetches live MLS data, reads the photos, writes the analysis, ships the report, and takes corrections by email. It runs as part of the realtor Agent Assistant, on a stack of n8n, LLMs, Cloud Run, and Supabase.
Book a free call. You will leave with a clear plan for your own AI automation, whether or not we work together.
Book a free call