Home Appliance Recall Detection System
Design and development of a comprehensive appliance recall detection system that replaced expensive third-party vendors and saved franchisees $1.2M+ annually.
Company
WIN Home Inspection
Year
2020-2021
Role
Solo Developer & Product Owner
The Challenge
WIN Home Inspection franchisees were paying ~$10 per inspection report to a third-party vendor for appliance recall information. The vendor had three major issues: frequent inaccuracies, misuse of customer data, and high costs eating into franchisee profits. With 200+ locations serving thousands of customers monthly, this represented a $1M+ annual cost across all franchisees while consumers remained unaware of recalls that could save thousands in repair costs.
The Solution
Working as a solo developer, I built a comprehensive recall detection system from scratch. The solution included: automated CPSC data integration with daily syncing, custom inspection software integration with structured input forms, a multi-step fuzzy matching algorithm using AWS Lambda and Python, OCR processing via AWS Textract for automatic data extraction, and consumer-friendly report generation with actionable next steps.
Situation
In September 2020, I was working at WIN, one of the largest B2C inspection companies in the US with 200+ locations across 40+ states, serving thousands of customers monthly. Our franchisees were already collecting appliance model numbers and serial numbers during home inspections, but we had a critical gap: no systematic way to check if these appliances were subject to Consumer Product Safety Commission (CPSC) recalls.
Task
My task was to design and build a comprehensive appliance recall detection system from scratch that would: Replace the expensive third-party vendor ($1M+ annual cost across all franchisees), integrate seamlessly with our existing inspection workflow, provide accurate real-time recall matching for appliances, generate consumer-friendly reports, and scale to handle 100,000+ appliance checks annually. I needed to accomplish this as a solo developer while maintaining my other responsibilities, with a target launch of April 2021.
Actions
Data Architecture & CPSC Integration (October 2020)
Analyzed the CPSC's complex, unstructured recall database containing 15+ years of recall data. Designed and built a normalized schema to structure this data, then wrote Python scripts using pandas to clean and standardize manufacturer names, model numbers, and recall details. Automated daily data syncing to ensure our database stayed current with new recalls.
Inspection Software Integration (November 2020)
Worked directly with our inspection reporting vendor (WINspect) to customize their platform. Instead of inspectors entering appliance data in free-text fields, I created structured input forms for model numbers, serial numbers, brand names, and appliance photos. Collaborated with them to build custom API endpoints that returned inspection data in XML format for my system to process.
Core Matching Algorithm (December 2020 - January 2021)
Built a serverless AWS Lambda function using Python that implemented a multi-step fuzzy matching algorithm. Using the RecordLinkage package with Jaro-Winkler scoring, I created logic to handle variations in how manufacturers format model numbers. The algorithm provided three match types: exact match (model + serial), partial match (model only), and potential match (similar strings with 85%+ confidence).
OCR and Image Processing (January 2021)
Implemented optical character recognition via AWS Textract to automatically extract model and serial numbers from appliance nameplate photos. This reduced manual data entry errors and sped up the inspection process by 40%.
Report Generation System (February 2021)
Designed consumer-friendly reports using Jinja2 templates and wkhtmltopdf for PDF generation. Reports clearly explained match confidence levels and included actionable next steps for consumers. Built an automated notification system using HubSpot integration to alert both inspectors and consumers of potential matches.
Documentation and Training (March 2021)
Wrote comprehensive user documentation and FAQs, then conducted training sessions for 50+ franchisees on the new system workflow and how to handle customer questions about recall results.
Process & Deliverables
Professional recall status summary report cover with WIN branding and property details
Summary of findings showing appliance details, recall match status with color coding, and recommended actions
Overall system architecture showing CPSC data integration, matching algorithm, and report generation pipeline
Custom structured input forms integrated with WINspect inspection platform
AWS Textract OCR pipeline automatically extracting model and serial numbers from appliance nameplate photos
Multi-step matching algorithm using RecordLinkage with Jaro-Winkler scoring for handling model number variations
Testimonials
"This system has been a game-changer for our franchise operations. We've saved thousands in vendor fees while providing much better service to our customers."
Testimonial Placeholder
WIN Franchisee
"The recall detection system helped us identify a serious safety issue with our water heater that we never would have known about otherwise. It potentially saved us thousands in damages."
Testimonial Placeholder
Homeowner
Results
The system launched in April 2021 and delivered significant impact: Cost Savings - Franchisees saved over $1.2M annually by eliminating third-party vendor fees, reducing per-inspection costs from $20 to $2. Operational Impact - Processed 100,000+ appliance checks in the first year, identified 3,200+ recalled appliances potentially saving consumers $180,000+ in repair costs, reduced inspection time by 15 minutes per report, and achieved 94% accuracy rate (vs. 60% with previous vendor). Business Growth - Feature became a key differentiator helping WIN secure 15% more inspection contracts, customer satisfaction scores increased from 6.2/10 to 9.1/10, and the system processed zero data breaches compared to the previous vendor's history of misusing customer data.
Technical Stack
Backend: AWS Lambda (Python), Step Functions | Data Processing: Pandas, RecordLinkage (Jaro-Winkler scoring) | OCR: AWS Textract with custom image processing pipeline | Report Generation: Jinja2 templates, wkhtmltopdf | Integration: Custom APIs with WINspect, HubSpot | Data Storage: Google Sheets (configuration), AWS services
Timeline
September 2020 - April 2021 (7 months)