Generative AI for Driftwood Capital

Client Introduction

Driftwood Capital is a vertically integrated commercial real estate firm specializing in hospitality investments. Their work involves evaluating large volumes of hotel acquisition and development opportunities. Each opportunity typically arrives as an email containing one or more PDF prospectuses, detailing:

• Room counts
• Parking capacity
• Market comps
• Amenities
• Financial projections
• Architectural and operational notes

Driftwood’s investment analysts must manually extract and evaluate all this information before it enters their investment pipeline.

Problem / Client Challenges

Despite receiving high-value opportunities daily, Driftwood faced several operational bottlenecks:

1. Massive Inbox Volume

Dozens of inbound hotel offering emails arrive each week, each containing multiple PDFs ranging from a few pages to over a hundred.

2. Manual Data Extraction

Analysts had to manually read PDFs to extract structured details (e.g., number of rooms, parking spaces, amenities). This consumed hours per opportunity.

3. Unstructured, Inconsistent PDF Formats

Prospectuses varied widely in design, format, and terminology. No two brokers deliver deals the same way.

4. No Centralized, Queryable Data Store

Investment data was not consistently normalized or stored in a warehouse for:

• Instant comparative analysis
• Automated underwriting
• Trend identification across past deals

5. Desire for Secure, Scalable AI (No Shadow IT)

Driftwood required an enterprise-secure solution inside their AWS environment, not a third-party AI vendor.

Solution: A Secure, End-to-End RAG Pipeline on AWS

• AllCode designed and delivered a production-grade RAG (Retrieval-Augmented Generation) pipeline purpose-built for Driftwood’s workflow, fully contained within their AWS environment.

Solution Overview

1. PDF → JSON Extraction Pipeline

All inbound PDFs are:

1. Uploaded to Amazon S3
2. Processed via a custom extraction pipeline using:
   - OCR (for scanned PDFs)
   - LLM-based document parsing
3. Transformed into normalized JSON with fields such as:
   • total_rooms
   • parking_spaces
   • meeting_space_sqft
   • brand
   • year_built
   • renovation_history
   • asking_price

2. RAG (Retrieval-Augmented Generation) Layer

To allow analysts to query all past deal documents, we built a modular RAG system:

• Amazon Bedrock Knowledge Base for vector embeddings & retrieval

• LangChain / CrewAI for agentic orchestration

• Swappable LLMs (Anthropic Claude, Amazon Titan, OpenAI GPT)

3. Secure API Layer for Internal Use

We deployed a hardened API surface:

• FastAPI

• Hosted on ECS Fargate

• Fronted by an Application Load Balancer with IP whitelisting

• Full HTTPS enforcement

4. Data Warehouse Storage in Snowflake

The structured JSON outputs are normalized and loaded into Snowflake, enabling:

• BI dashboards

• Instant deal comparisons

• Automated valuation calculations

• Internal predictive modeling

5. Automated CI/CD

Delivered via:

• GitHub Actions

• AWS CDK (Python) for Infrastructure-as-Code

• Blue/green deployments to ECS

This ensures rapid iteration without risk.

Results

50% Reduction in Analyst Processing Time

Automated extraction cut time spent parsing PDFs in half.

📄 Standardized, Queryable Deal Data

All deal attributes now flow into Snowflake in structured JSON form—enabling analytics
never before possible.

🔐 100% Compliance with Security Requirements

No data ever leaves Driftwood’s AWS environment.

🔥 Production-Ready in 4 Weeks

We delivered:

• Infrastructure

• Ingestion

• RAG pipeline

• API

• CI/CD

• Testing

…all within a single month.

💬 Non-Technical Users Can Ask Natural Questions

Examples:

• “Compare parking ratios across all Hilton-branded deals submitted this quarter.”

• “Which Florida hotel opportunities included renovation plans since 2015?”

• “Show me all opportunities with more than 300 rooms and significant group demand.”

📈 Better Decision-Making

Driftwood now uses AI-augmented insights to:

• Pre-screen deals faster

• Eliminate non-viable opportunities early

• Focus analyst time on high-value underwriting

• Strengthen competitive advantage in the hospitality investment market

Conclusion

Driftwood Capital transformed a high-volume, manual, error-prone process into a fully automated, secure, scalable AI pipeline using AWS-native services.

By integrating RAG, structured data extraction, Snowflake warehousing, and enterprise security, AllCode delivered a solution that saves hundreds of analyst hours per year and unlocks deeper insights into hotel investment opportunities.

This case study demonstrates the power of combining cloud-native architecture with applied GenAI to create real, measurable business outcomes.