AEGIS-CORE :: Neural Industrial Orchestrator

Bridging the Multi-Decade Gap Between Legacy Hardware and Agentic AI

---

🌟 The Problem: The "Siloed" Industrial Reality

In modern manufacturing and energy plants, the most critical data is often trapped. It lives in legacy SQL databases from the 90s, proprietary sensor networks without APIs, and "black-box" PLC systems.

Solving issues in these environments usually requires manual data entry, cross-referencing printed logs, and human operators acting as the only "integration layer." This process is slow, error-prone, and disconnected from the rapid pace of modern decision-making.

AegisCore was built for the Amazon Nova Hackathon to change this. We provide the "Neural Bridge" that finally connects these legacy systems to the world's most advanced reasoning engines.

---

🛡️ Project Vision

AegisCore is a mission-critical Agentic AI framework designed to orchestrate industrial automation. By using Amazon Bedrock Nova 2 Lite, we don't just provide a chatbot—we provide a digital supervisor that can see into your databases, feel your hardware's pulse, and execute complex cross-platform workflows safely.

---

🧩 Technical Architecture

graph TD
    User((Plant Operator)) -->|Natural Language| Frontend[Neural Dashboard]
    Frontend -->|WebSocket| Bridge[Python Agent Bridge]
    
    subgraph "The Intelligence Layer (Amazon Bedrock)"
        Bridge -->|Stateful Converse API| Nova[Amazon Nova 2 Lite v1:0]
        Nova -->|Tool Call Reasoning| Bridge
    end
    
    subgraph "The Physical Layer (MCP)"
        Bridge -->|Grounding Request| MCPServer[Data Grounding Engine]
        MCPServer -->|Query| SQLDB[(Legacy SQL Database)]
        MCPServer -->|Telemetry| Sensors[Industrial Hardware]
    end
    
    Bridge -->|Final Actionable Report| Frontend

Loading

---

🚀 The Core Innovation: "Agentic Grounding"

1. The Reasoning Engine: Nova 2 Lite

We selected the us.amazon.nova-2-lite-v1:0 model for its exceptional ability to handle Converse API Tool Use. Unlike simpler models that hallucinate, Nova 2 Lite follows a strict "Observe-Orient-Decide-Act" (OODA) loop:

• Observe: Listens to the operator's natural language request.
• Orient: Identifies which legacy system (SQL or Sensor) has the answer.
• Decide: Formulates a structured tool call.
• Act: Interprets the raw hex/data results and translates them back to human terms.

2. The Model Context Protocol (MCP) Server

Grounding is everything. Our custom-built MCP Server acts as the translator. It converts legacy SQL rows and raw sensor voltages into "Clean JSON" that Nova can understand.

• Inventory Check: Cross-references browser-extracted data with the master SQL inventory.
• Hardware Telemetry: Simulates real-time polling of temperature and vibration signatures for Industrial Sensors.

3. Human-In-The-Loop (HITL) Safety

In industry, a wrong click can be catastrophic. AegisCore implements a Policy-Driven Gatekeeper. Even if the AI determines a write-action is necessary, the system pauses, presents the reasoning in the "Orchestrator Logs," and requires a physical human signature (Approval) before the bridge executes the command.

---

💬 Demo Guide: Sample Prompts

To experience the true agentic power of AegisCore, try these prompts in the dashboard:

Role	Operational Prompt	What the AI does behind the scenes
Logistics	"Cross-reference the browser data with the legacy database and check if we have enough Control Valve B in stock."	Extracts data, calls cross_reference_inventory, queries SQL, and summarizes the mismatch.
Engineer	"Provide a full system health report and specifically check the telemetry for Industrial Sensor A."	Calls get_system_status and get_sensor_telemetry, polls mock hardware, and generates a technical health diagnostic.
Safety	"Are there any discrepancies in the sensor vibration levels across the plant?"	Orchestrates multiple sensor checks to identify anomalies in the telemetry stream.

---

🏗️ Deep Dive: The MCP Layer

The Model Context Protocol (MCP) is the "Hands" of our AI. It allows the model to interact with the world through two primary tools defined in supervisor.py and handled by nova_act_mcp_server.py:

• cross_reference_inventory:

  • Inputs: List of objects {name: string, quantity: number}.
  • Logic: Performs an exact match against the legacy_inventory SQLite table.
  • Output: Returns a list of MISMATCH or NOT_IN_DB findings.

• get_sensor_telemetry:

  • Inputs: {sensor_name: string}.
  • Logic: Simulates hardware polling via a randomizer to provide realistic Temperature and Vibration flux.
  • Output: Returns a simulated hardware state JSON.

---

📁 Project Detailed Structure

├── backend/
│   ├── agents/
│   │   ├── agent_api.py      # The WebSocket Bridge (The "Spinal Cord")
│   │   ├── supervisor.py     # The Orchestrator (The "Frontal Lobe")
│   ├── Dockerfile            # Container definition for the Bridge
│   ├── requirements.txt      # Python dependencies (boto3, websockets)
├── frontend/
│   ├── src/
│   │   ├── components/       # Reusable Atomic UI Components
│   │   ├── App.tsx           # Dashboard Orchestration & State
│   │   └── index.css         # Global Industrial Design System
├── mcp_examples/
│   └── nova_act_mcp_server.py # The Translator between AI and Data
├── .gitignore                # GitHub exclusion rules
└── README.md                 # This technical manifesto

---

📦 Deployment & Containerization

Docker Deployment

AegisCore is ready for enterprise container environments. You can launch the backend bridge using:

cd backend
docker build -t aegis-bridge .
docker run -p 8765:8765 -e AWS_ACCESS_KEY_ID=XXX -e AWS_SECRET_ACCESS_KEY=XXX aegis-bridge

GitHub Readiness

The project includes a comprehensive .gitignore to ensure that:

• ❌ Secrets (CSV files, credentials) are never pushed to the server.
• ❌ Environment folders (venv, node_modules) are excluded.
• ❌ Artifacts (Gemini brain logs) remain local.

---

🛠️ Installation & Manual Getting Started

Prerequisites

• AWS Bedrock Access: Ensure us.amazon.nova-2-lite-v1:0 is enabled in your region.
• Python 3.10+ & Node.js 18+.

1. Launch the Neural Bridge (Backend)

cd backend/agents
pip install boto3 websockets
python agent_api.py

2. Launch the Dashboard (Frontend)

cd frontend
npm install
npm run dev

---

🛰️ Roadmap: Beyond the Hackathon

Our vision for the next 12 months includes:

• Edge Deployment: Running the Bridge on Greengrass devices.
• Computer Vision: Using Nova's multimodal capabilities to analyze thermal feeds.
• Regulatory Logs: Automated PDF reporting for safety audits.

---

Developed for the Amazon Nova Hackathon 2026. Empowering the future of industrial intelligence.