ValuAI | Devpost

Inspiration

Insurance companies often struggle to verify property damage claims, as contractors may overprice repair costs, leading to inflated payouts. Homeowners, on the other hand, may find it hard to prove the extent of their damages. We built ValuAI to ensure transparency by using AI to cross-verify claims, helping insurers detect exaggerated quotes and homeowners get fair assessments.

What it does

ValuAI helps insurance companies verify property damage claims by using multimodal AI systems. Users upload images of damages and provide repair estimates. Our AI processes both text descriptions and images simultaneously — detecting complex issues like sun damage to flooring — and flags suspiciously high quotes. This allows insurers to validate claims quickly and accurately.

How we built it

We used Flask for backend processing and integrated multimodal AI, combining text and image models. The text model repeatedly communicates with the image model to refine damage detection and repair cost estimation. We utilized LLaMA and Google Flask, connected via the OpenRouter API, enabling seamless AI interactions and real-time analysis.

Challenges we ran into

Training AI to detect both obvious and subtle damages, like structural cracks versus sun damage, required high-quality datasets and continuous model tuning. Ensuring effective collaboration between text and image models was another hurdle, as accurate, context-aware assessments depend on real-time communication between these models. Additionally, validating contractor quotes against realistic repair costs was challenging, as it required reliable data sources to spot inflated claims.

Accomplishments that we’re proud of

We successfully developed a multimodal AI system that cross-verifies damage claims, combining text and image analysis for better accuracy. We created a tool that not only helps insurance companies prevent overinflated contractor quotes but also supports homeowners by making the claim process more transparent. Integrating LLaMA and Google Flask through the OpenRouter API allowed us to implement real-time AI interactions, enhancing both speed and precision.

What we learned

Accurate damage verification requires more than just image analysis — context from text descriptions is crucial for AI predictions. We learned that AI models need continuous retraining with high-quality data to detect complex damages, such as sun discoloration or hidden structural issues. Additionally, real-time collaboration between text and image models significantly improves accuracy and reduces errors, reinforcing the need for seamless multimodal AI integration.

What’s next for ValuAI

We plan to build RAG (Retrieval-Augmented Generation) databases to compare claims against historical data, allowing for more accurate predictions and better fraud detection. By integrating APIs, we aim to pull average repair costs and cross-reference claims with past records. Additionally, we want to develop a mobile app for on-the-go claim verification, making the process faster and more accessible. Partnering with insurance companies will help streamline claim processing and prevent inflated contractor quotes. Our main challenges include finding useful datasets to train and validate our AI models and managing API request limits while ensuring real-time data retrieval for precise damage assessments.