💡 Inspiration
We built SafeHouse after realizing how many people live in unsafe housing conditions without even knowing it. Problems like mold, water leaks, and faulty wiring often go unnoticed until they become serious health risks.
Traditional home inspections are expensive, slow, and inaccessible to most people. We wanted to create something that makes housing safety instant, preventive, and understandable using AI.
🏡 What it does
SafeHouse is an AI-powered web application that analyzes images of a living space to detect potential health, safety, and accessibility risks such as mold, leaks, structural damage, unsafe electrical setups, and environmental hazards.
It goes beyond detection by turning findings into clear, actionable support:
🛠️ Repair recommendations tailored to each identified issue
📍 Nearby repair shop suggestions, including top-rated options
🛒 Essential product recommendations via Amazon for fixing or improving the space
🧠 Simple, non-technical explanations of risks and what to do next
♿ Neurodivergent-friendly insights that reduce cognitive overload through clear, structured guidance
SafeHouse transforms any home into a fast, accessible, and supportive safety check system designed to help people feel more secure, informed, and in control of their environment.
⚙️ How we built it
We built SafeHouse as a modular AI system:
🖼️ Image input pipeline for analyzing home environments
🤖 Computer vision model for detecting visual hazards
🧠 LLM layer to convert detections into human-readable insights
📊 Risk scoring system to prioritize issues by severity
🌍 Location-based service lookup for repair shops
🛒 Recommendation system for essential home products
The architecture is designed to be extensible for real-time scanning in future versions.
⚠️ Challenges we ran into
Translating raw visual detections into meaningful, user-friendly insights
Mapping detected issues to real-world repair categories
Balancing multiple recommendations without overwhelming the user
Designing a system that remains fast and lightweight in a web environment
Ensuring outputs are both accurate and actionable
🏆 Accomplishments that we're proud of
Successfully combining computer vision + LLM reasoning into one system
Turning detections into actionable real-world recommendations
Adding repair + product + local service intelligence in one workflow
Building a full end-to-end working AI prototype in a short hackathon timeframe
Focusing on preventive health through housing safety
📚 What we learned
How to connect AI perception with decision-making systems
The importance of translating technical outputs into human impact
Designing AI tools for prevention rather than reaction
Building practical UX for health-related AI systems
Structuring multi-agent style pipelines (vision + reasoning + retrieval)
🚀 What's next for SafeHouse
📱 Add real-time camera-based scanning for live home inspection
🧠 Improve detection accuracy with fine-tuned vision models
🗺️ Smarter geo-aware repair and service recommendations
🏠 Add “move-in safety score” for apartments and rentals
🧑⚕️ Integrate environmental health risk predictions
🤝 Partner with rental platforms and housing marketplaces
Log in or sign up for Devpost to join the conversation.