ArgoAI

I'm building an agentic AI underwriting tool for a New York City real estate development firm. Anonymized client work.

The Problem

Real estate underwriting is time-intensive — analysts gather data from dozens of sources, build financial models, and synthesize it all into investment recommendations. A single deal can take days to evaluate properly.

What It Does

A chat-based agent that helps the firm assess whether a given build site is worth pursuing. The agent:

• Searches the web for live market, regulatory, and comparable-deal context as the conversation unfolds
• Performs RAG retrieval over the firm's internal corpus of memos, zoning research, and prior diligence
• Aggregates structured data from real estate APIs (parcels, comps, market metrics)
• Learns from the conversation itself — the analyst can drop additional source documents into the chat mid-conversation and the agent incorporates them into the working assessment

The output is a working assessment, not a finished underwriting model. Judgment stays with the analysts; the agent removes the data-gathering tax.

Stack

FastAPI backend with LangGraph for agentic orchestration. React frontend with Leaflet for map visualizations. PostgreSQL with pgvector for RAG retrieval over the firm's internal documents. Tool use integrates web search (Tavily), real estate APIs, and live document uploads.