System & AI Architect | GPU Inference | Cloud & Distributed Systems

AI & Machine Learning Architectures
Expertise in Deep Learning, Sequence Models (RNNs, LSTMs for NLP), CNNs, Reinforcement Learning, MLOps Pipelines, Model Evaluation Frameworks, as well as LLM and Prompt Engineering Foundations. Formal training through the Deep Learning Specialization (Coursera, Andrew Ng) with a grade of 96.67%.

GPU Inference & LLM Serving
Hands-on experience commissioning and optimizing bare-metal NVIDIA HGX H200 inference servers, deploying large language models via vLLM with tensor/expert parallelism, building highly available LLM gateways using LiteLLM, HAProxy, keepalived/VRRP, Patroni/etcd PostgreSQL clusters, Redis Sentinel, and observability stacks with DCGM/Prometheus/Grafana. NVIDIA-Certified Associate in both AI Infrastructure & Operations and Generative AI LLMs.

Cloud & Distributed Systems
Proven track record architecting and deploying high-performance distributed systems on the Microsoft Azure Ecosystem, including Azure Batch, Compute Fleet, Scale Sets, Databricks, Data Factory, and App Services. Orchestration of compute fleets with up to 8,000 vCPUs, leveraging Docker, Kubernetes (Polyaxon, Kubeflow), and Service-Oriented Architecture (SOA).

Engineering & Automation
Strong background in Data Pipeline Engineering, Workflow Automation, Numeric Determinism, High-Performance Computing, Algorithm Optimization, and CI/CD Templates. Designed custom post-parallel ordering deterministic reduction strategies to guarantee bit-wise identical numerical results across varying compute scales.

Programming Languages & Tools
Proficiency in C#, C/C++, Python (TensorFlow, PyTorch, Scikit-learn, Pandas, NumPy), MATLAB, JavaScript/TypeScript, SQL, Grafana, and Power BI. Applied across production pipelines, algorithm development, and data science workflows.

Reinforcement Learning & Multi-Agent Systems
Architecture of robust Reinforcement Learning agents for complex multi-agent environments, including end-to-end data pipelines, feature engineering layers, action masking, distributed training paradigms, and custom evaluation frameworks.

MLOps & Model Deployment
Design and launch of end-to-end scalable MLOps and machine learning data pipelines, transitioning experimental models into production-ready containerized solutions using MLflow, Polyaxon, and Kubeflow on bare-metal cluster servers.

Transfer Learning & Computer Vision
Application of transfer learning and advanced supervised deep learning to architect high-performance models for automated optical inspection, achieving significant reduction of false positives while maintaining strict recall targets.

Bare-Metal Provisioning & Platform Tuning
Experience with MAAS bare-metal provisioning, BIOS and platform settings tuning for GPU throughput (NUMA/NPS1, IOMMU passthrough, ACS, Resizable BAR, power and C-state configuration), and Ansible/GitOps deployment automation.