Lead AI/ML Engineer & Data Scientist | Production ML & GenAI | Distributed Systems | GCP · Azure · GPU Computing

Lead AI and ML engineer and data scientist with 10+ years of experience, building production AI/ML/LLM systems. Leads end-to-end AI delivery, combining hands-on model development with technical leadership to turn business problems into scalable, high-throughput, low-latency production systems. Experience spans predictive modelling, forecasting, operations research, computer vision, and generative AI across retail, cybersecurity, manufacturing, insurance, and supply-chain optimization.

Skills

- Agentic AI & LLMs: Google ADK; Gemini / Vertex AI; Tool Calling; Agent Evaluation; LoRA / QLoRA; OpenAI-Compatible Serving

- Modelling & Analytical Methods: Classical ML; Deep Learning; Computer Vision; NLP; Forecasting; Feature Engineering; Operations Research; Autoencoders

- Frameworks & APIs: scikit-learn; XGBoost; PyTorch; TensorFlow; Hugging Face / PEFT; Unsloth; vLLM; FastAPI; ONNX Runtime; OR-Tools

- GPU & Distributed Computing: Multi-GPU Training; RAPIDS cuDF / CuPy

- Model Lifecycle: MLflow; Evaluation / Registry; Data / Model Quality Gates; Reproducible Training; Vertex AI / Azure ML Pipelines; Feature Store; Semantic Retrieval

- Cloud & Deployment: GCP / Azure; GKE / Kubernetes; Docker; Cloud Run / Functions; Azure Functions / Logic Apps; Helm / Istio

- Reliability & Reporting: CI/CD; Cloud Build; Argo CD / GitOps; Prometheus / Grafana; OpenTelemetry / Cloud Trace; GitHub Actions / pytest; Load Testing; Streamlit / Power BI

- Languages, Data & Storage: Python; SQL; Beam / Dataflow; Pub/Sub; BigQuery; Redis; GCS / Azure Blob; Pandas / NumPy; Dask; Git / Linux

Experience

Lead AI/ML architecture and end-to-end delivery across LLMs, search, recommendations, fraud and cybersecurity.

- Designed and built an autonomous root-cause analysis agent as sole developer with Google ADK, Gemini on Vertex AI and Python. Investigates production incidents using metrics, logs, traces, Kubernetes state and source code, with scoped tool access, read-only guardrails and evidence-backed reports; reduces investigation time from hours to minutes.

- Productionized a reusable LLM fine-tuning framework with Unsloth, PEFT, Hugging Face Transformers and Accelerate, supporting LoRA/QLoRA, distributed training, task-specific adapters and automated evaluation.

- Deployed vLLM inference on GKE with OpenAI-compatible endpoints, HPA autoscaling, Istio routing and Prometheus/Grafana monitoring, with estimated annual API-cost savings of ~$300K.

- Lead ML engineering for sponsored-product recommendations handling nearly 3 million requests/day: automated feature/training pipelines, model-quality gates, semantic retrieval and serving on GKE across two regions. Set architecture and release standards with data science, application, cloud and reliability teams; introduced reusable AI coding-agent skills and automated checks.

- Built retail fraud feature pipelines at 40,000 events/sec at peak using Beam/Dataflow and Vertex AI Feature Store. Improved holiday score coverage by ~20% and deployed XGBoost through FastAPI on Kubernetes; the model's contribution saves US$1.5M/year.

- Engineered AWS account digital fingerprinting on NVIDIA DGX (8 x A100) using PyTorch autoencoders and RAPIDS cuDF/CuPy on billion-row datasets, with continuous retraining, MLflow and Grafana alerts.

- Scaled GPU-accelerated phishing detection to 2,400 emails/sec (2x throughput); integrated analyst feedback and Microsoft Sentinel for detection and quarantine.

- Automated ML/OCR-based abuse-inbox triage, doubling case volume. Mentored engineers, coordinated client/NVIDIA delivery, screened 300+ resumes and conducted 50+ technical interviews.

- Developed product-quality prediction for BamaGruppen with fully automated training and serving using Logic Apps and Azure ML platform services, including Azure ML pipelines, model registry and container deployments.

- Increased truck inspections from 130 to 160/day, defect detection by 17% and customer approval by 20%, delivering $2.5M annual savings.

- Replaced manual Excel planning with automated last-mile routing and transport optimization using OR-Tools and mixed-integer models, integrating Azure Functions, Logic Apps and RPA to enable rapid replanning and improve delivery efficiency.

- Modelled supply-chain and distribution-hub scenarios across cost, freshness, carbon and warehouse volumes using OR-Tools on Azure ML with Power BI reporting.

- Led cross-functional delivery, mentored data scientists and presented network-design recommendations to Norwegian stakeholders.

- Developed forecasts and scenario analysis for GlobalData across 15+ industry verticals with forecasting and feature-engineering pipelines, informing market intelligence and resource allocation.

- Developed semiconductor defect detection with TensorFlow / Keras DenseNet on a GCP multi-GPU cluster. Distributed training on ~1 million images in ~3 hours achieved 90% classification accuracy.

- Ingested SQL Server data and served inference with Docker / TensorFlow Serving. Deployed the end-to-end system at Amkor, South Korea, and trained the client team.

- Implemented opportunity win/lose prediction with logistic-regression scoring and data pipelines for sales prioritization; integrated SQL Server Integration Services, SSAS and Power BI dashboards.

- Delivered tested insurance ID-card workflows using COBOL, JCL, DB2, VSAM and CICS; handled impact analysis, design and support, and shared knowledge with the team.

Selected Recognition

Best Buy Talent Quotient (2026) - cybersecurity and fraud delivery.

Best Buy client AML Hall of Fame (2024), during the Wipro engagement - DGX framework delivery and offshore mentoring.

Accenture Delivery Excellence; Kudos x2.

Certifications

- Google Cloud Certified: Generative AI Leader | Microsoft Certified: Azure AI Engineer Associate

- Microsoft Certified: Azure Data Scientist Associate | Microsoft Certified: Power BI Data Analyst Associate

- Oracle Cloud Infrastructure 2024 Generative AI Certified Professional

Education