Instead of relying on human operators to stare at monitors and match anomalies across disparate systems, OmniSight automatically validates, deduplicates, and enriches these events in real time. Once an anomaly is detected, an Agentic Workflow kicks in to handle the cognitive heavy lifting.
1. Executive Summary
OmniSight (Autonomous Multi-Modal Incident Intelligence Fabric) is an advanced, enterprise-ready AI platform designed to ingest and correlate high-velocity signals from diverse edge sources—including RTSP Cameras, IoT Sensors, Microphones, and Standard Operating Procedure (SOP) Documents.
It replaces the traditional "alert fatigue" model with an autonomous investigator. It retrieves semantic evidence from vector memory, reasons through risk factors, generates safety guardrails, and presents a holistic view via a high-end, premium Operator Dashboard built on FastAPI.
2. Core Use Cases
OmniSight is designed to be highly adaptable to various physical-world monitoring scenarios:
- 🏭 Smart Manufacturing & Industrial IoT: A critical CNC machine exhibits an IoT temperature spike while an audio sensor detects a grinding anomaly. OmniSight correlates the two, queries the machine's maintenance manual using RAG, and issues a severe alert recommending shutdown.
- 🏢 Warehouse Safety & Logistics: A camera detects a forklift moving at high speed in a pedestrian-only zone. OmniSight logs the safety violation, links it to the safety protocol document, and automatically drafts an incident report for the shift supervisor.
- 🏙️ Smart City Operations: Traffic cameras detect a collision, and nearby microphones register a loud impact. OmniSight fuses the multi-modal data, assesses severity, and prepares a structured payload for emergency dispatch APIs.
- 🔐 Secure Facility Monitoring: A badge-swipe is registered at a secure door, but the camera detects two faces (tailgating). OmniSight flags the anomaly, cross-references access control policy, and alerts the guard station.
3. Agentic Workflow Engine
OmniSight moves beyond simple rules-based alerts by utilizing a Deterministic Agentic Runtime composed of specialized AI personas, ensuring autonomous action without hallucination:
The Personas
- 👁️ Observer: Continuously monitors the event stream and identifies raw anomalies (e.g., "Temperature > 85°C").
- 🕵️ Investigator: Groups related anomalies within specific time-windows and locations into structured Incidents.
- 📚 Retriever: Queries the Qdrant/Postgres vector database to find relevant SOPs, safety manuals, or past resolutions.
- 🗺️ Planner: Analyzes the evidence and formulates a step-by-step resolution plan.
- 🛡️ Safety Guard: A deterministic policy engine that reviews the Planner's suggestions to ensure no autonomous actions violate enterprise safety rules.
- ⚙️ Action Executor: Dispatches approved actions (e.g., sending emails, creating Jira tickets, triggering webhooks).
4. Architecture & Data Flow
The backend is powered by FastAPI, processing JSON payloads at high velocity. The Agent Runtime handles the LLM logic, heavily utilizing RAG to fetch localized safety documents.
Cameras/IoT] -->|JSON Payloads| B[FastAPI Gateway] B -->|Schema Validation| C[Event Processor] C -->|Time-Window Match| D[Incident Correlator] D --> E[Memory Fabric] E -.->|Vector Search| F[Semantic DB] E -.->|State| G[PostgreSQL] D --> H[Agent Runtime] H -->|RAG| F H -->|Policy Check| I[Safety Guard] I --> J[Action Execution] I --> K[Operator Console] style A fill:#0f172a,stroke:#38bdf8,stroke-width:2px,color:#fff style B fill:#1e293b,stroke:#3b82f6,stroke-width:2px,color:#fff