home proof ai

Inspiration

Each year, millions of families lose their homes to floods, fires, or storms — and most can’t prove what they owned. Over 92% of households lack an up-to-date inventory, which delays or reduces insurance payouts. We wanted to fix that with technology that sees, reasons, and documents — while keeping data private and explainable.

Our idea: What if you could walk through your home for 10 minutes and have a complete, insurer-ready dossier generated automatically? That question inspired HomeProof AI — an agentic system that turns visual context into structured, auditable evidence.

What it does

HomeProof AI builds a claim-ready home inventory in minutes using on-device vision and AWS Bedrock agentic workflows. It turns quick room scans into insurer-grade reports with confidence scores.

How we built it

A Next.js PWA captures and processes frames locally with ONNX/PyTorch Mobile. AWS Lambda, SQS, and EventBridge orchestrate a CrewAI agent lineup on Bedrock AgentCore, with Neo4j as the knowledge graph.

Challenges we ran into

Coordinating on-device ML with serverless Bedrock agents, managing spiky workloads, and maintaining explainability under strict privacy and cost constraints.

Accomplishments that we're proud of

End-to-end working demo: 15-minute home inventories, 90%+ item accuracy, full audit trail, and explainable valuations — all within a $0.50/item processing cost.

What we learned

Agentic workflows shine when each agent’s role is clear and data context is shared. Combining event-driven AWS infra with reasoning agents gives both scale and transparency.

What's next for home proof ai

Extending to insurer APIs, multi-language interfaces, and fine-tuned reasoning models for appraisal and damage detection — bringing trust and speed to real-world claims.

Built With

• agentcore
• bedrock
• crewai
• eventbridge
• lambdas
• mcp
• neo4j
• nextjs
• onnx
• python
• pytorch
• s3
• serverless
• sqs
• typescript