Back to Blog
January 2025

Enhancing Roof Damage Assessment and Reporting for Insurance Claims with RAG

This blog presents a roof damage assessment and reporting solution powered by the AMD Instinct MI300X accelerators on Dell PowerEdge XE9680 servers.

| Industry Challenges in Roof Damage Assessment

Insurance companies often face significant challenges in assessing and reporting roof damage efficiently. Traditional workflows rely heavily on manual inspections and detailed reporting, leading to time-consuming processes that delay claim submissions and resolutions. These inefficiencies not only frustrate policyholders but also increase operational costs and introduce inconsistencies in claim evaluations. With thousands of claims requiring physical inspections annually, the lack of automation hampers transparency, consistency, and speed, straining customer satisfaction and insurer resources.

| A High-Performance Solution

To address these challenges, we developed an automated Roof Damage Assessment and Reporting solution powered by Dell PowerEdge XE9680 servers equipped with AMD Instinct MI300X accelerators. This solution transforms the insurance claims process by automating damage detection and streamlining report generation. Integrating advanced vision language models (VLMs), audio multimodal models, and retrieval-augmented generation (RAG), it enhances operational efficiency and customer experiences.

This blog delves into the solution architecture and highlights its capabilities:

  • Automated detection and classification of roof damage with RAG and multimodal models.
  • Deployment of advanced models on Dell PowerEdge XE9680 servers, utilizing AMD Instinct MI300X accelerators.
  • Rapid inspection report generation in compliance with industry standards to reduce claim processing delays and improve transparency.

| Solution Architecture

To power this solution, we selected the Dell PowerEdge XE9680 equipped with AMD Instinct MI300X accelerators due to its exceptional performance and memory capacity. With 192GB of HBM3 memory per accelerator, we can comfortably run the entire Llama 3.1 70B model on a single accelerator.

This software stack includes the following key components:

  • rocm-vLLM v0.6.4, an industry-standard library for optimized open-source large language model (LLM) serving, with support for AMD ROCm 6.2.
  • Llama 3.1 70B Model, an industry-leading open-weight language model with 70 billion parameters.
  • Molmo 72B VLM, an industry-leading vision language model served on vLLM.
  • Ultravox, a multimodal language model that processes audio and text for real-time voice interactions.
  • EVF SAM, an advanced model that enhances text-prompted segmentation.
  • bge-large-en embeddings model, one of the top ranked text embeddings models on Hugging Face APIs.
  • MilvusDB, an open-source vector database with high performance embedding and similarity search.
  • LightRAG, a retrieval-augmented generation (RAG) model that uses graph structures and dual-level retrieval.

| Solution Overview

This solution automates key aspects of damage identification, area segmentation, and report generation, ultimately reducing manual effort, accelerating claims submissions, and enhancing transparency and consistency.

VLM-Powered Damage Identification Process:
The core of this solution is the Molmo 72B Vision-Language Model (VLM), which processes high-resolution images collected during roof inspections. The model identifies and describes key types of roof damage, including flashing damage, shingle damage, and chimney damage.

Integration with Inspector Feedback (Audio):
To enhance the assessment's accuracy, the solution integrates inspector feedback via transcribed audio recordings, processed using the Ultravox 8B Audio Multimodal model.

Automated Marking of Damage Areas:
The EVF-SAM (Segment Anything Model) automatically segments the images to mark areas of roof damage.

Automated Roof Damage Report Generation:
Using Retrieval-Augmented Generation (RAG), the solution generates a comprehensive damage report for each home, customized to include information from the homeowner's policy document.

Real-Time Dashboard for Hardware Performance:
The solution features a dynamic UI dashboard that showcases hardware performance metrics.

By integrating state-of-the-art AI models and scalable hardware, this solution revolutionizes the roof inspection process, significantly reducing the time and effort required for damage assessments and claims submissions while enhancing overall accuracy and transparency.

To learn more, please request access to our reference code by contacting us at contact@metrum.ai.

Copyright © 2024 Metrum AI, Inc. All Rights Reserved. This project was commissioned by Dell Technologies. Dell and other trademarks are trademarks of Dell Inc. or its subsidiaries. AMD, AMD Instinct, AMD ROCm, and combinations thereof are trademarks of Advanced Micro Devices, Inc. All other product names are the trademarks of their respective owners.

DISCLAIMER - Performance varies by hardware and software configurations, including testing conditions, system settings, application complexity, the quantity of data, batch sizes, software versions, libraries used, and other factors. The results of performance testing provided are intended for informational purposes only and should not be considered as a guarantee of actual performance.

Back to All Articles