I specialize in building and scaling Autonomous AI-powered IT Operations (AIOps) platforms and Agentic Architectures (A2A). My work focuses on transitioning machine learning and multi-agent systems from local sandbox notebooks into resilient, microservices-driven production deployments capable of managing high-frequency telemetry at scale.

graph TD
    %% Styling Definitions
    classDef primary fill:#1e1e2e,stroke:#cba6f7,stroke-width:2px,color:#cdd6f4;
    classDef secondary fill:#181825,stroke:#89b4fa,stroke-width:1px,color:#cdd6f4;
    classDef data fill:#1e1e2e,stroke:#a6e3a1,stroke-width:1px,color:#a6e3a1;
    
    %% Graph Structure
    Ingest[Telemetry Stream: 140k+ Routers] -->|FastAPI Ingestion| Buffer(Kafka Buffer / Queue)
    Buffer -->|Log Processing| Factory{Agent Factory}
    
    subgraph A2A Orchestrator Core
        Factory -->|Session 1| WiFi[Invincible WiFi Agent]
        Factory -->|Session 2| Modem[Cable Modem Agent]
        Factory -->|Session 3| Offload[Mobile Offload Agent]
    end
    
    WiFi & Modem & Offload -->|Action Validation| IaC[Terraform ECS Fargate]
    WiFi & Modem & Offload -->|Diagnostics| DB[(DynamoDB / Neptune Graph)]

    class Ingest,Factory primary;
    class WiFi,Modem,Offload secondary;
    class Buffer,DB,IaC data;

An autonomous AIOps telemetry pipeline and multi-agent self-healing orchestration console managing nationwide dual-link router networks.

• Architectural Milestones:
  • A2A Inter-Agent Comm: Implemented agent_card.json schema discovery mapping telemetry alerts directly to specialized sub-agents.
  • High-Performance Ingestion: Migrated legacy AWS Lambda functions to direct FastAPI REST backend, dropping request latency from 1200ms to <150ms.
  • Isolated Session Contexts: Designed a dynamic factory pattern preventing session cross-contamination across concurrent dashboard engineers.
  • Enterprise Tooling (MCP): Converted monolithic utility files into modular @tool schemas querying DynamoDB and Aurora databases.
  • Infrastructure as Code: Provisioned all ECS Fargate, ALB, and IAM resources via modular Terraform scripts.
• Technologies: Python 3.11 | FastAPI | React | Terraform | Docker | Apache Kafka | Neptune

Industrial AI platform for predictive maintenance using sensor data, Remaining Useful Life (RUL) prediction, OEE metrics, and ML models.

• Architectural Milestones:
  • Feature Engineering: Engineered rolling standard deviations, sensor differences, and temporal windows on noisy time-series telemetry.
  • Model Comparison: Built and optimized Recurrent Neural Networks (LSTM/GRU) and gradient-boosted ensembles to perform reliable predictive regression.
  • Environment Standardization: Structured production-grade environment management using PyProject/Pipenv ensuring 100% pipeline reproducibility.
• Technologies: Python | TensorFlow | Scikit-Learn | Pandas | MLflow | Docker

All projects are validated using a custom automated compliance framework. Here is a typical output from my local CI suites:

[TEST] STARTING STANDALONE AIOPS VERIFICATION TESTS

-> Test 1: Verifying Pydantic Request validation...
   [OK] Valid device ID passed successfully.
   [OK] Invalid device ID format correctly raised validation error.

-> Test 2: Verifying Reusable Tool definitions and schemas...
   [OK] Tool decorators and schemas verified successfully.

-> Test 3: Verifying Factory Pattern user isolation...
   [OK] Agent Factory isolated user contexts correctly.

[SUCCESS] ALL LOCAL VERIFICATION TESTS PASSED SUCCESSFULLY!