# SAF-Integration - **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon) - **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST) - **Where:** Online - **Team:** [Brad Wentzel](https://cerebralvalley.ai/u/SAF-Integration) - **GitHub:** https://github.com/BoMa-AI-Cloud/aria-home - **Demo video:** https://youtu.be/ZqxiowDCa0g - **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery - **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1367 ARIA Home is designed to help older adults live independently while giving their families greater peace of mind. It checks essential household systems, identifies potential problems, and helps designated caregivers respond before minor issues become serious. Its purpose is to provide an additional layer of oversight, especially overnight or when family members cannot be physically present. I'm an electrician based in Petoskey, Michigan, and I've seen firsthand the extensive damage that can result when household problems go undetected. As my in-laws get older, I've become increasingly concerned about what might happen if their heat failed in the middle of the night and nobody noticed until morning. ARIA Home was developed to address situations like this. By bringing together information from independent sources, including room sensors, thermostats, electrical activity, calendars, weather forecasts, and cameras, ARIA can recognize when something isn't working as expected. Rather than simply sending another alert, she evaluates the situation, determines when caregiver involvement is needed, and helps coordinate an appropriate response with the caregiver's approval. Our demonstration focuses on one practical question: Will the home reach 68°F by 7:00 AM, and if not, who should be notified to help resolve the problem? In our simulated demonstration, the home begins the evening warm and on track. At 9:41 PM, the room temperature drops to 66°F, triggering a notice that the home may not reach its morning target. At this stage, ARIA records the concern without contacting anyone. By 10:46 PM, the temperature has fallen to 62°F, prompting ARIA to ask the designated family caregiver a single question: "May I contact the HVAC company?" These 2°F and 6°F differences are configurable alert thresholds, not established safety limits. An additional safeguard prompts a response if the home remains at risk within two hours of its scheduled temperature goal. The demonstration evaluates these conditions at predetermined times. ARIA does not contact the HVAC company without approval. In the simulated scenario, the caregiver authorizes the call at 10:47 PM, and the resident receives a notification that a technician may arrive. The technician reaches the home at 11:30 PM, just 44 minutes after ARIA requests permission, and completes the repair at 12:15 AM. The room temperature then begins to rise. If the caregiver declines the request, no service provider is contacted. The home remains cold, and ARIA documents that the morning temperature goal was not met. Importantly, ARIA does not assume the problem is resolved simply because the thermostat reconnects or a technician reports completing the repair. Instead, she verifies the result using independent room-temperature readings sustained for 30 minutes. In the successful scenario, the goal is confirmed at 1:41 AM, less than three hours after the initial request. The system's code determines when action is needed, verifies outcomes, and controls who receives notifications. Meta's models assist with explanations and reviews but do not make operational decisions. You can explore the working demonstration here: https://aria-home-demo.onrender.com The public demo uses synthetic data and simulated messages, requires no sign-in, and does not control actual equipment. ARIA Home is not intended to serve as a medical or emergency response service. ARIA Home was developed during the Meta Global AI Developer Hackathon, with all code available in the project repository. Based on Git history, Meta's models generated approximately 95% of the code, while Claude by Anthropic contributed at least 5%, as documented in docs/evidence/code-authorship.md. Under my direction, Claude developed the 2°F and 6°F alert rules, the resident's technician-visit notification, and the final wording revisions. These contributions are recorded in docs/build-log/operator-log.md. The accompanying video is a dramatization of the application's simulated scenario, with that distinction clearly identified on screen. Its scenes were generated using Meta AI, and certain elements, including a concert in Honolulu, were added for storytelling purposes. The live demonstration allows viewers to experience the actual application and its decision-making process. The narration, including the lines in my first person, is a synthetic voice (Kokoro-82M), used only to make the video. The application never calls it. ## More from Meta Global AI Developer Hackathon - [Orbit Fitting Room](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1364) - [RepoLens](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1365) - [Glint](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1366) - [ALMA PROJECT TEAM](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1368) - [John Stamford](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1369) - [Vastraa Labs](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1370) --- Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1367. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt