Industries:
Software and Internet
Regions: 
NA
AWS Segment: 
Startup

The Challenge

As California experiences an increasing number of wildfires, insurance companies have tightened their property coverage requirements. In response to this situation, the State of California Department of Insurance issued new guidelines, REG-2020-00015, aimed at promoting sustainable development, reducing greenhouse gas emissions, and requiring accurate assessments of flammable materials near insured structures. Climup Tech saw a need for a robust and automated AI/ML solution to identify fire hazards, meet the new insurance requirements, and comply with the broader goals of this regulation. This includes adhering to energy efficiency standards, supporting renewable energy promotion, and demonstrating a commitment to waste management and public awareness initiatives.

Why ClimUp Chose Cloud303?

Expertise in AI/ML Solutions: Cloud303 is a leading AI/ML solution provider with extensive experience in developing robust, scalable AI/ML systems. Their proficiency in using AI/ML technologies was crucial in crafting an innovative solution for Climup Tech.

Strong AWS Partnership: Cloud303's strong partnership with AWS meant that they had a deep understanding of AWS's services and how to use them effectively. This partnership ensured that the solution they developed for Climup Tech was optimized for the AWS environment.

Proven Track Record: Cloud303 had a history of successful deployments and satisfied clients. This gave Climup Tech confidence in their ability to deliver a solution that met their specific needs and compliance requirements.

Innovative Approach: Cloud303 demonstrated an innovative approach to problem-solving. Their use of cutting-edge technologies like AI/ML and depth estimation models like AdaBins showcased their ability to tackle complex tasks effectively.

Project Management and Collaboration: Cloud303's professional approach to project management and collaboration ensured that all project objectives were met within the required timeline, while maintaining high standards of quality and compliance.

Cloud303's Solution

Cloud303 collaborated with Climup Tech to develop a comprehensive AI/ML solution, leveraging several AWS services to create models and classifiers, manage data and files, and provide functional APIs for client interaction. The solution encompassed state-of-the-art computer vision and analysis techniques, and made extensive use of AWS AI/ML services for an engaging and elaborate experience.

Cloud303 deployed Amazon SageMaker, a fully managed service that enables developers to build, train, and deploy machine learning models quickly and efficiently, for various aspects of the project.

SageMaker was used to develop various AI/ML models and classifiers, including InExClass (for image classification), image segmentation, depth estimation (using AdaBins), and FireHazardAdjacency (for detecting flammable items within five feet of a structure). AWS Rekognition was employed for image segmentation, while AWS Textract was used for text parsing from vendor-provided data. Climup Tech also implemented semantic segmentation using recipes from AWS, and created house bounding roof-lines from images for better hazard assessment.

SageMaker was also employed to support data labeling, driving towards an auto-labeling capability for the initial product training. This helped create a high-quality, labeled dataset to train the AI/ML models.

InExClass, an AI/ML classifier, was built on AWS Rekognition and Python to determine whether an image is an exterior image or some other type (e.g., interior, floor plan, general outdoor). This classifier played a critical role in filtering relevant images for further analysis.

To estimate 3D depth from 2D images, Cloud303 utilized Amazon SageMaker to deploy a pre-trained AI/ML model, AdaBins for this specific task. The AdaBins algorithm inputs a single 2D image and predicts a depth map.  This was augmented with clever programming for localized depth comparisons, as AdaBins provides an imperfect depth map.

By integrating this model into the system, Climup Tech was able to gain better insights into the spatial relationships between the insured structure and nearby flammable objects. This added depth information significantly improved the system's ability to assess the proximity of flammable materials and determine the potential fire risk.

SageMaker was also used to create FireHazardAdjacency, an AI/ML classifier, or algorithm, to determine if flammable objects identified by the segmentation model were within five feet of the structure. This helped in assessing fire hazards in close proximity to the insured structures.

Cloud303 developed server-side code using serverless AWS Lambda to cost-efficiently support the on-demand nature of the workloads.  AWS Fargate is then applied to both queue and spin up elastic compute resources to process the thousands of properties per API call in a timely manner.  Each Fargate task uploads aerial and street view images from Google or Bing in addition to other real estate photos from separate image providers, invokes Sagemaker endpoints for image segmentation, depth analysis, and performs fire hazard adjacency analyses on the results.

Climup utilized AWS Textract, a machine learning service that automatically extracts text and data from scanned documents, to create or use a text parser to gather detailed information about a property as provided by the vendors. This parser extracted the relevant data from scanned documents, which was then saved to the database for further analysis and cross-referencing with the image analysis results.

Amazon S3 and Amazon RDS were utilized to create an organized file and database structure to maintain original and processed images, metadata, interim results, and other relevant information.

The integrated system enabled users to input property addresses, gather images from vendor APIs, and automatically apply the AI/ML models for fire hazard proximity assessment.

Results/Benefits

The AI/ML solution, jointly developed by Climup Tech and Cloud303, successfully enabled Climup Tech to assess fire hazards to enable insurance companies to comply with the new insurance requirements. The system's high accuracy, automation, and scalability made it a valuable tool for insurance companies and their clients.

By leveraging AWS services like Amazon SageMaker, AWS Rekognition, and AWS Textract, Cloud303 was able to showcase their AI/ML expertise and build a robust, efficient, and scalable solution withClimup Tech. This collaboration enabled Climup Tech to help CA insurance companies to meet the stringent insurance requirements; it has positioned Climup Tech as pioneers in adopting cutting-edge AI/ML technologies for property risk assessment.

Furthermore, the system has enabled Climup Tech to attract insurance companies as potential clients. The solution's comprehensive features and seamless integration with existing workflows have made it attractive for insurance providers looking to automate and streamline their fire hazard assessment processes.

The partnership between Climup Tech and Cloud303 has successfully deployed an innovative AI/ML solution using AWS services to address the challenge of assessing the proximity of flammable materials to insured structures. The solution has not only helped Climup Tech clients comply with new insurance guidelines but has also enhanced Climup Tech’s reputation as a technology-driven company in the property management and insurance space.

With Cloud303's expertise in AI/ML and the power of AWS services, Climup Tech is now better equipped to tackle the challenges of a dynamic insurance landscape while offering improved services to their clients.