Back to projects
AIProductReal Estate

Lease Intelligence

Self-serve Q&A over 880 commercial leases

Jun – Aug 2026Empire State Realty Trust, New York, NY

Overview

Lease teams at ESRT answered lease questions the slow way: find the lease, read the lease, ask legal to confirm. Lease Intelligence is the prototype I built to make that self-serve: ask a question in plain English about any of 880 leases and get an answer with citations back to the source document.

It has two halves. Ask is the conversational side: Claude synthesizes cited answers over abstracts produced by Harvey-managed retrieval. Dashboard is the structured side: letter-of-credit expirations, rent steps, renewal windows, and options surfaced across the portfolio, with a browsable abstract for every lease.

I presented the prototype and its cost case to company leadership at the end of the summer.

Sandbox demo

The actual app, running entirely in your browser on a small fabricated portfolio. Every tenant, address, and figure is invented. Ask a question and get a cited answer, or open the dashboard and click into a lease for its full abstract.

sandbox demo · dummy data · no backendOpen full screen ↗

The problem

Two manual workflows ate the team's time: producing lease abstracts (reading and summarizing every executed lease) and answering one-off lease questions that each required pulling the underlying document. Neither scaled across an 880-lease portfolio.

Approach

  • Automated abstraction with Harvey Agents and Playbooks: RAG-based narrative abstracts plus tabular abstracts verified against Yardi, cutting abstraction time roughly in half.
  • Built Ask: a conversational interface where Claude synthesizes answers over Harvey retrieval, with citations back to the exact lease language.
  • Built Dashboard: portfolio-wide views of LOC expirations, rent steps, renewals, and options, plus a per-lease abstract browser.
  • Shipped the prototype as a FastAPI + React app on Azure App Service behind Entra ID single sign-on.

Outcomes

  • Modeled an 86% licensing-cost reduction versus per-seat Harvey licenses ($227.3K vs $4.9–32.8K per year).
  • Cut lease abstraction time by roughly 50% with Harvey automations.
  • Covered all 880 leases in the portfolio with searchable, cited abstracts.
  • Presented the prototype and rollout plan to company leadership.

Stack & tools

Claude (Anthropic API), Harvey Agents + Playbooks, FastAPI, React, Azure App Service, Entra ID, Yardi.

Next projectCANDI