RE Alpha Deal Intelligence Engine

RE Alpha Engine combines document intelligence, financial modeling, and live market intelligence to autonomously underwrite institutional real estate deals

Comment

Inspiration

Institutional real estate investing is still surprisingly manual.

Acquisition teams spend hours parsing Offering Memorandums (OMs) from platforms like LoopNet and Crexi, rebuilding financial models in Excel, researching submarket comps, pulling macro data, and drafting investment memos — all before an IC (Investment Committee) even looks at the deal.

The insight behind RE Alpha was simple:

What if an AI agent could function as a junior acquisitions analyst — instantly extracting, modeling, stress-testing, and contextualizing a deal within its live market environment?

As someone passionate about AI-native workflows in high-leverage industries, I wanted to build something that felt like a real institutional tool — not a toy demo.

What it does

RE Alpha is an AI-powered deal intelligence engine for institutional real estate teams.

Upload an OM (PDF), and the system:

Extracts structured deal data (rent roll, purchase price, NOI, cap rate, assumptions)

Reconstructs financial projections

Calculates IRR and equity multiple

Runs scenario and sensitivity analysis

Identifies risk factors

Pulls live market intelligence

Generates an investment verdict memo

Instead of reading 60 pages and building a model from scratch, users receive:

A structured deal snapshot

Modeled base case and stress case returns

Key risk flags

Market context (comps, trends, macro signals)

A clear Buy / Watch / Pass recommendation

It behaves like an AI acquisitions associate — available instantly.

How we built it

RE Alpha is a multi-agent, vertically integrated AI system.

Document Intelligence Layer

Reka.ai for structured extraction from OMs

Claude Vision for qualitative interpretation and risk reasoning

Financial Modeling Layer

Custom IRR + scenario engine

Cash flow modeling based on extracted assumptions

Sensitivity logic (rent growth, exit cap expansion, vacancy stress)

Market Intelligence Layer

Tavily for live web search

Submarket trends, comparable transactions, macro signals

Context injected directly into the investment thesis

Agent Orchestration

Claude as the primary reasoning engine

Structured prompts, rules, and skill files to simulate disciplined underwriting logic

Output structured into IC-ready memo format

Frontend

Lightweight upload interface

Rendered summary + financial outputs

Clean investment memo view for demo clarity

The goal was not just to “analyze a document” — but to create a full-stack AI investment workflow.

Challenges we ran into

  1. Financial precision vs. LLM creativity Large language models are excellent at reasoning — but underwriting requires discipline. We had to separate deterministic calculations (IRR engine) from narrative reasoning (Claude) to avoid hallucinated math.

  2. OM variability Every OM is formatted differently. We solved this using structured extraction schemas and fallback logic.

  3. Signal vs. noise in market search Live search can easily introduce irrelevant information. We designed structured search prompts and constrained outputs to ensure signal quality.

  4. Time constraints Building document extraction, modeling, market intelligence, and deployment within hours required tight architectural discipline and ruthless scope control.

Accomplishments that we're proud of

Accomplishments that we're proud of

Fully working end-to-end underwriting pipeline

Structured OM extraction from real deal PDFs

Automated IRR + scenario modeling

Live market intelligence integration

Investment memo generation that feels institutional

Clean, demo-ready UI

Most importantly:

This feels like a real product that an acquisitions team could actually use.

What we learned

Vertical AI > general AI. Deep domain structure dramatically improves output quality.

Deterministic + probabilistic systems must be separated in finance workflows.

Live market context makes AI decisions significantly more persuasive.

Agents are most powerful when scoped as “roles” (e.g., junior acquisitions analyst) rather than generic assistants.

Institutional workflows are ripe for AI automation.

What's next for RE Alpha Deal Intelligence Engine

Portfolio-level intelligence

Track multiple deals

Rank by risk-adjusted return

Compare capital allocation scenarios

Neo4j-based relationship graph

Sponsor track record mapping

Broker network intelligence

Historical deal performance linking

Automated IC Deck Generation

PowerPoint-ready investment memos

Waterfall visualizations

Real-time pipeline monitoring

Monitor submarkets continuously

Alert when new deals meet investment criteria

Institutional integrations

CRM sync (Juniper Square, Salesforce)

Data room auto-generation

Capital stack structuring

Long-term vision:

Become the AI-native operating system for institutional real estate acquisitions.

Built With

Share this project:

Updates