Commercial Real Estate · Document AI✓ POC — Demo Successful
AI-Generated Letters of Intent for Commercial Real Estate
Built a POC that automated the drafting of commercial real estate Letters of Intent — taking deal specifications from a structured input and intelligently filling a bracketed LOI template to produce a complete, ready-to-review document.
The Problem
Drafting a Letter of Intent for a commercial lease deal is tedious and error-prone. Every LOI uses the same template, but each deal has different economics, timelines, parties, and terms. The manual process meant copy-pasting values, hunting for every bracketed placeholder, and inevitably missing fields or inserting inconsistent language.
A third-party vendor was looking to bring this capability to their CRE clients and needed a working POC to take to market.
The Solution
The system took a bracketed LOI template and deal specifications from a structured Excel input, then used a multi-step pipeline to index every placeholder in the document, understand its surrounding context, and fill it accurately.
Rather than doing a dumb find-and-replace, it used RAG to understand what each field meant in the context of the full document before filling it — handling ambiguous or context-dependent placeholders intelligently. Output was a complete, formatted LOI document ready for attorney review.
Pipeline
01
Index
All bracketed placeholders in the LOI template are identified and indexed with their surrounding sentence context.
02
Parse
Deal specifications are extracted from the structured Excel input — economics, parties, dates, lease terms — and normalized.
03
Embed
Template chunks are embedded into a vector store. Each placeholder's context is used to retrieve semantically relevant deal spec fields.
04
Resolve
For each placeholder, the pipeline retrieves relevant context and determines the correct value — handling ambiguous or compound fields intelligently.
05
Fill
Placeholders are replaced with resolved values. The output is a complete, formatted LOI document with all fields populated.
Clause Coverage
Tenant & LandlordPremises AddressLease TermCommencement DateRent TypeBase RentRent EscalationsSecurity DepositPermitted UseExclusivityTI AllowanceOptions to Renew
Outcome
The POC was built, demoed, and landed well — the vendor confirmed the output quality met their bar for client presentation. The engagement stalled when their end customer decided not to move forward, a business decision unrelated to the product. The system validated the core approach: context-aware placeholder resolution produces significantly better output than template-filling tools, especially for deals with non-standard terms.
Tech Stack
LLMGPT-4o
FrameworkLangChain
Vector DBChromaDB
InputExcel deal spec + .docx template
OutputFilled LOI .docx
Have a similar problem? Let's talk.
Book a free call →