Project Overview
Liberty Settlement Funding (Liberty SF) needed a faster, more accurate way to mine U.S. court orders for prospects who may wish to sell future structured-settlement payments. Avahi designed and deployed LibertySF Entity Extraction—an end-to-end, event-driven pipeline that ingests court-order spreadsheets, extracts more than 25 legal and financial entities with Amazon Bedrock’s Nova Pro large-language model, and returns a readyto-use Excel lead list. The solution cuts processing time from hours to minutes per batch while improving data quality and scale.
About the Customer
Liberty Settlement Funding is a specialty finance firm headquartered in Fort Lauderdale, Florida. The company purchases structured-settlement payment rights and annuities from individuals who need immediate liquidity.
The Problem
Liberty SF’s business development team previously reviewed thousands of court orders manually—a slow, errorprone process that limited the number of prospects they could contact and delayed time-to-offer. As filing volumes climbed, manual review threatened to bottleneck growth, increase labor costs, and allow competitors to reach claimants first.
Why AWS
Liberty SF chose AWS for its managed, highly scalable services and built-in generative-AI capabilities:
- Amazon Bedrock provided fully managed access to state-of-the-art foundation models such as Nova Pro without infrastructure overhead.
- Event-driven services (Amazon S3, EventBridge, ECS Fargate, RDS) offered elastic scale, secure data isolation inside a private subnet, and pay-as-you-go pricing that aligns with variable filing volumes.
Why Liberty SF Chose Avahi
As a Premier-tier AWS Partner specializing in GenAI, Avahi brought proven cloud architecture patterns, Bedrock expertise, and an agile delivery model that guaranteed production launch in six weeks. Avahi’s legal-document accelerators and deep knowledge of AWS funding programs further reduced Liberty SF’s upfront risk and cost.
Solution
- Ingestion & Orchestration – Users drop an Excel docket into an Amazon S3 “source” bucket. An EventBridge rule triggers an ECS Fargate task that orchestrates the workflow.
- Document Retrieval – The task iterates through each spreadsheet row, downloads the linked HTML Page from S3, and streams content to Amazon Bedrock.
- Entity Extraction with Nova Pro – Bedrock’s Nova Pro model identifies 25+ fields (case number, minors, payment streams, settlement amounts, legal flags, etc.) with high precision, returning structured JSON. The model ID is managed centrally in AWS Parameter Store.
- Persistence & Output – Extracted entities are saved to Amazon RDS (MySQL). When all rows are processed, the service generates a new Excel file, stores it in an S3 “destination” bucket, and notifies users for download.
- Persistence & Output – Secrets Manager secures credentials; CloudWatch captures logs and metrics; ECR hosts the processing image; the entire stack resides in a private subnet for data privacy.
Key Deliverables
- Automated ETL pipeline (S3 → EventBridge → ECS → Bedrock → RDS)
- Nova Pro-based entity-extraction framework
- Excel lead-list generator with criteria-based filtering (e.g., exclude divorce or mortgage cases)
- Versioned data storage and backup in S3/RDS
- CloudWatch monitoring & runbooks
- Deployment scripts and hand-over documentation
Project Impact
Avahi’s solution gave Liberty SF an always-on, zero-touch pipeline that scales with filing volumes and eliminates manual data entry.
Measured results
- Processes ~4 court orders per minute—16 seconds per document—from upload to finished spreadsheet
- Handles large batches (e.g., 6,000 records ≈ 26 hours) with predictable throughput
- Extracts 25+ legal and financial entities with model-level accuracy, enabling rapid, targeted outreach
