Project snapshot
By modernizing its aerial imagery analysis system on AWS, a property intelligence innovator has reduced costs by 10x while scaling its property data processing capabilities. This transformation allows insurers and property managers to quickly and accurately assess property risks and values, delivering peace of mind for thousands of property decisions every day.
Client
Arturo.ai, a leading property intelligence company based in the United States, specializes in advanced aerial imagery analytics for property assessment. Their AI-based platform helps insurance companies, property owners, and managers make informed decisions by offering clear, evidence-based insights into property conditions.
Business challenge
The client faced major challenges with their batch-based, resource-intensive ML pipeline, which struggled to meet the increasing demand for timely data processing needed to fulfill SLAs. To address urgent production issues, they required immediate fixes to stabilize operations, along with a plan for long-term pipeline optimizations, including transitioning from batch processing to a more scalable, real-time approach.
The critical operational challenges included:
- Legacy batch-based ML pipeline unable to scale with business growth
- Resource-intensive processing leading to high operational costs
- Limited capability to process real-time property data
- Complex infrastructure management reducing focus on core business operations
- Need for enhanced security and compliance in property data processing
Technology solution
Our collaboration began with a two-week technical assessment service focused on MLOps optimization. To resolve critical production challenges, we implemented a proof of concept (POC) that uses AWS SageMaker’s latest model deployment techniques, balancing cost savings and performance for a complex cascade of eight computer vision models.
As the next step, our team built an AWS-based serverless solution architecture with the following components:
Model deployment and inference comprising:
- multiple property feature and roof condition models using AWS SageMaker multi-model endpoints
- integrated model versioning and registration through SageMaker Model Registry
Data processing pipeline consisting of:
- aerial-captured geospatial data, which is processed and enriched using a machine learning pipeline, then stored in Amazon S3 and DynamoDB
- event-driven preprocessing using AWS Lambda functions and orchestration by AWS Step Functions
- real-time inference processing through SageMaker Endpoints
- results storage and notification system using Amazon SNS
Security and compliance part integrating:
- comprehensive IAM roles and policies following least privilege principles
- multi-factor authentication (MFA) for enhanced security
- AWS CloudTrail for complete API activity monitoring
- encrypted data transmission using TLS and AWS KMS
- AWS Shield and Web Application Firewall (WAF) for advanced threat protection
Monitoring and operations featuring:
- CloudWatch monitoring for model accuracy, endpoint latency, and resource utilization
- automated alerts for performance metrics through CloudWatch Alarms
- comprehensive runbooks for routine operations and troubleshooting
- automated testing through AWS CodeBuild and CodePipeline
Disaster recovery initiatives establishing:
- RTO of 30 minutes and RPO of 1 hour
- cross-region replication for critical data
- automated recovery processes using CloudFormation StackSets
Business outcomes
Our collaboration with a client demonstrates the impact of MLOps in transforming an industry-specific machine learning pipeline into a highly efficient, scalable, and future-ready system. Together, we achieved measurable technological and operational improvements:
- Cost optimization: Achieved a 10x reduction in total cost of ownership (TCO) through strategic architectural redesign
- Operational efficiency: Transformed machine learning infrastructure to accelerate property analysis and enable rapid market adaptation
- Enhanced security: Implemented advanced technological solutions that strengthen data security and system reliability
- Technological agility: Developed a resilient ecosystem with comprehensive monitoring and robust security controls
- Performance gains: Improved system responsiveness and streamlined operational processes
Through this transformation, the client positioned themselves for continued growth in the property intelligence market, balancing immediate cost savings with long-term strategic advantages.