Skip to content
NR
00/Case Study

CRE Intelligence Platform

NYC Agentic Underwriter & Feasibility

RoleCreator
Timeline2026
StackReact, FastAPI, Python
SourceGitHub
Impact

Autonomous commercial real estate underwriting platform mapping NYC comparable sales and zoning limits, dropping feasibility report latency to ~4 seconds.

Overview

Context

The CRE Intelligence Platform is an autonomous, source-cited commercial real estate underwriting and feasibility tool for NYC brokers and developers. It instantly resolves properties, queries live public tax registries, evaluates zoning limits, maps comparable sales, and underwrites development scenarios.

Challenge

The problem

Evaluating commercial real estate feasibility in NYC typically requires manually pulling data from multiple disparate sources (PLUTO, ACRIS, Zoning Resolution) and running offline spreadsheet calculations. The challenge was to create a unified platform that automates this workflow, providing instant, accurate, and visual underwriting with real-time agentic insights.

Approach

How I built it

01

Implemented a speed-optimized parallel pipeline resolving addresses via DCP GeoSearch, fetching DoITT PLUTO lots, and querying ACRIS Deed Sales via asyncio

02

Integrated a Zoning RAG vector database to retrieve district explanations

03

Developed a 3D buildable massing visualizer using pure CSS 3D transforms (zero WebGL overhead)

04

Built an interactive pro forma calculator for both rental residential and condo sellout scenarios

05

Created a comparable sales heatmap overlay using Leaflet and an 'Ask the Agent' chat drawer powered by Claude Sonnet

Technical Decisions

Why these choices

Speed-Optimized Parallel Pipeline

Using asyncio.gather to fetch PLUTO and ACRIS data in parallel drastically reduces wait times, enabling single-turn Claude evaluations and dropping report latency to ~4 seconds, avoiding Vercel serverless timeouts.

Pure CSS 3D Massing Visualizer

Leveraging CSS 3D transforms for lot massing eliminates the need for heavy WebGL libraries, ensuring the application remains lightweight, fast, and responsive across devices.

Single-turn Claude Analysis

Feeding compiled records and zoning context to Claude in a single turn provides rapid feasibility synthesis, avoiding multi-step agent latency while keeping CRE insights concise and direct.

Outcomes

What shipped

Interactive 3D massing visualizer with real-time footprint and story height scaling
Dynamic pro forma underwriting calculator for rental and condo scenarios
Side-by-side site comparison with local storage persistence
Live production deployment on Vercel: realestate-agentic.vercel.app
Takeaways

What I learned

Parallelizing external API calls is critical for serverless environments with strict timeout limits
Pure CSS can achieve surprisingly effective 3D visualizations without the overhead of WebGL
Providing raw data context directly to LLMs enables highly specific and useful real estate insights