ClaimGuard

Inspiration

The idea for ClaimGuard was inspired by the need to streamline and enhance the insurance claims process. Traditional methods of assessing property damage, detecting fraud, and calculating compensation are often time-consuming, error-prone, and inefficient. By leveraging advancements in artificial intelligence (AI), machine learning (ML), and real-time data analysis, ClaimGuard aims to revolutionize this process, making it faster, more accurate, and transparent for both insurers and claimants. The goal is to reduce fraudulent claims while ensuring fair compensation for genuine cases.

What it does

ClaimGuard is an AI-powered system that simplifies and automates the property insurance claims process. It performs the following functions: • Damage Assessment: Analyzes legal documents, photos, videos, and interviews to determine the extent of property damage. • Fraud Detection: Identifies inconsistencies in claims using AI-driven analysis of documents, media, and interviews. • Real-Time Cost Estimation: Calculates repair costs dynamically based on geographic location, market prices, and event-specific factors (e.g., regional disasters). • Compensation Recommendation: Suggests tailored payouts and bonuses based on insurance policies, historical data, and the current status of the affected area. • Actionable Insights: Provides real-time dashboards with damage severity scores, cost estimates, fraud risk levels, and recommendations for next steps.

How we built it

Our tech stack is centered around an event driven architecture on AWS. We have four major parts to our application: a frontend, an s3 bucket that stores our application’s multimodal inputs, a custom trained ai model that’s running using Microsoft’s ONNX runtime in rust, and a frontend made using react and tailwind. When the insurance agent makes a claim on our dashboard, they upload their images, interviews, and pdf documents to an s3 bucket using an s3 pre-signed url. This triggers a lambda that subscribes to the bucket. Based on the type of the uploaded image, this lambda routes the input to one of our five different ai models. Three of these models are Anthropic API calls, which return a response which is saved to the database. For our emotion model, instead of a directly coupled API call to our model, we pass our inputs through an SQS queue. This allows us to process long video calls in parallel, increasing our application speed immensely. Our frontend is relatively simple, and made using react, tanstack router, and tailwind css.

Challenges we ran into

One challenge with using the Anthropic API was formatting the answer in a way that made it easy for us to use. Additionally, deploying our custom trained tensorflow model to AWS was a big challenge for us. Our aws framework's python support wasn't working. Because one of us had experience with deploying a model using Rust, we decided to go with that. To do this, we had to add rust bindings to the AWS framework, and use a local build of the framework during development. We also contributed it back to the framework as a pr (https://github.com/sst/sst/pull/5432).

Accomplishments that we're proud of

We are proud of the following accomplishments with ClaimGuard:

AI-Powered Automation: Successfully developed an AI-driven system that automates the complex and often tedious process of property damage assessment, fraud detection, and cost estimation, significantly reducing manual effort and increasing accuracy.

Scalability: Our solution is designed to handle large volumes of multimodal inputs (documents, images, videos, and interviews) from various sources, ensuring it can scale as the insurance industry grows and adapts to new challenges.

Improved Fraud Detection: Through deep learning and sophisticated model training, we've built an effective fraud detection system that is capable of spotting inconsistencies across multiple data types, enhancing the integrity of the claims process.

Faster Processing: By leveraging event-driven architecture and AWS services like Lambda, SQS, and ONNX, we have streamlined the backend processes, allowing faster data processing and claim resolution.

What's next for ClaimGuard

The future roadmap for ClaimGuard includes: 1. Enhanced Fraud Detection: • Incorporating deeper cross-modal analysis to identify inconsistencies between text narratives and visual evidence more effectively. 2. Scalability Improvements: • Expanding the system’s capacity to handle larger datasets from multiple regions or insurance providers. 3. User Experience Enhancements: • Adding features like automated claim submission feedback or preventive recommendations based on property risk assessments. 4. Integration with Insurance Providers: • Partnering with insurers to deploy ClaimGuard as a core component of their claims management systems. 5. Advanced Analytics: • Leveraging predictive analytics to identify high-risk areas or properties prone to damage before incidents occur.