Abstractly：AI 驱动的商业地产租约提取与租金表校验工具

AI-powered lease abstraction and rent roll validation for commercial real estate.

Abstractly turns unstructured lease PDFs into clean, structured, source-cited data — then goes a step further by validating that data against rent rolls to catch the discrepancies that cost CRE teams time and money. What used to be hours of manual abstraction and line-by-line rent roll reconciliation becomes a workflow measured in minutes, with every extracted figure traceable back to the page and clause it came from.

Key Features

AI-powered lease abstraction — extracts tenant, rent, escalations, CAM charges, renewal options, and other key terms from lease PDFs, with a page-level source citation and confidence score on every field
Rent roll validation — imports a rent roll and automatically reconciles it against abstracted lease data, flagging mismatches before they become costly errors
Portfolio dashboard — aggregate metrics (total rent, CAM exposure, square footage), sortable/filterable lease list, and a portfolio health score at a glance
Automated risk detection — flags below-market rent, missing standard clauses, notice-period outliers, and inconsistent terms
Grounded Q&A assistant — ask plain-English questions about the portfolio and get answers backed by real citations, not guesses
Lease comparison & benchmarking — compare leases side by side or against the portfolio average
Exports — rent roll (CSV/Excel), investment memos, and printable portfolio summary reports
Multi-user access — role-based accounts (owner/admin) with secure authentication

Tech Stack

Backend: Python, Flask, SQLite — PDF/OCR parsing (pypdf, pdfplumber, Tesseract), Claude (Anthropic API) for AI-driven extraction and the Q&A assistant, bcrypt-based authentication

Frontend: Vanilla JavaScript, HTML5, CSS — no framework or build step

Deployment: Dockerized backend deployed on Render, static frontend served separately

How It Works

Upload — drag in one or more lease PDFs, plus a rent roll if you have one
Extract — Abstractly reads each lease and pulls out the key terms, citing the exact page and quote behind every value
Validate — rent roll figures are automatically checked against the abstracted lease terms, and any discrepancies are surfaced immediately
Analyze — the portfolio dashboard rolls everything up into metrics, risk flags, and a health score
Ask & export — query the portfolio in plain English, or export a rent roll, investment memo, or summary report

Demo

Screenshots / live demo link coming soon.

Getting Started

# Backend
cd backend
python3 -m venv venv && source venv/bin/activate
pip install -r requirements.txt
python run.py            # runs on http://localhost:5000

# Frontend
cd frontend
python3 -m http.server 8080   # open http://localhost:8080

AI-powered extraction and the Q&A assistant require an ANTHROPIC_API_KEY; without one, the app runs on its built-in rule-based extraction engine.

Abstractly is a portfolio project built to explore AI-assisted document intelligence for commercial real estate.