# Dristhi Cloud Muse APP - **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:** [Dhiraj Patra](https://cerebralvalley.ai/u/dhirajpatra) - **GitHub:** https://github.com/dhirajpatra/Drishti-Cloud - **Demo video:** https://youtu.be/6sCm1VGKOHI - **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery - **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/302 Hello judges, I'm Dhiraj Patra, and this is Drishti Cloud — Multi-Agent Superintelligence for Every Building. Imagine a security agency responsible for hundreds of cameras across homes, schools, shopping malls and hospitals. Human guards cannot continuously monitor every screen. Critical incidents can easily go unnoticed. Drishti addresses this problem by transforming ordinary cameras into intelligent, context-aware security assistants. Architecture and technology What makes Drishti different is its multi-agent architecture, powered by Meta's AI technologies. We use SAM 3.1 for visual segmentation, Muse Spark for multimodal scene understanding and decision-making, and Muse Wonder for contextual reasoning. These specialized agents communicate through Redis streams, coordinated by a FastAPI gateway. This is not a single LLM prompt. It is a distributed intelligence pipeline where every agent has a specific responsibility. Multi-tenancy and multimodal intelligence Drishti supports multiple independent tenants and security environments. A single platform can simultaneously manage a senior citizen's home, a shopping mall, a school and an ICU. Each tenant has isolated data, configurations, safety rules and memory. We combine camera images, sensor inputs, object segmentation, OCR and site-specific safety SOPs to understand what is happening rather than simply detecting objects. Live dashboard and intelligent alerts Here is the live monitoring dashboard. We can see multiple sites, their cameras, event statistics and real-time notifications. Drishti classifies incidents into NORMAL, ALERT and CRITICAL. For example, a potential fall, an unattended bag, physical violence or an ICU monitor alarm can trigger an appropriate response. Every alert includes its reason and an annotated evidence image, allowing a human operator to understand why the system raised it. Contextual reasoning and memory One of our key innovations is contextual intelligence. Using Neo4j, Drishti maintains a separate habit graph for each tenant. For example, if an elderly person normally takes an afternoon nap, the system can learn that behaviour and reduce unnecessary false alarms. Muse Wonder also considers recent events and local news to contextualise and escalate alerts. The system never allows learned habits to suppress critical emergencies such as physical violence. Actions, scalability and engineering When an incident is confirmed, the action agent can notify family members, security personnel or hospital staff through WhatsApp Business Cloud API. It also supports edge-device actions such as voice announcements and OLED notifications. We have implemented Docker Compose orchestration, PostgreSQL row-level security, Redis streams, Neo4j graph memory, model fallbacks and cost-control mechanisms. The system avoids unnecessary model calls when camera scenes remain unchanged. Our automated test suite currently includes 149 passing tests. Drishti is designed to make intelligent security accessible, scalable and affordable. It is a working implementation with real camera integration, Meta model integration and an offline simulator for reproducible demonstrations. Our vision is simple: One intelligent edge device can protect a home. A network of them can help protect an entire world. Thank you. ## More from Meta Global AI Developer Hackathon - [MuseOps](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/299) - [Guardian](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/300) - [Kataku.ai](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/301) - [CK Kang](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/303) - [Donovan Ratefison](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/304) - [Kaagaz_India](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/305) --- Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/302. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt