Vj W

AI Engineer

FedEx

Oct 2025 - Mar 2026 • 5 mos

• Architected and led development of scalable agentic AI systems integrating autonomous decision-making across a hybrid cloud environment (AWS Lambda, S3, Textract + Azure AI Services), demonstrating cross-cloud orchestration judgment in ambiguous enterprise settings. • Designed and implemented an orchestrator-executor multi-agent architecture using LangChain, applying ReAct and Plan-and-Execute patterns for autonomous operational decision-making; benchmarked Azure AI Agent Service vs. LangGraph for multi-agent coordination across cloud boundaries. • Shipped a Natural Language-to-SQL pipeline (Azure OpenAI + LangChain) enabling employees to query billions of rows of operational data in plain English and receive validated SQL output, delivered end-to-end in under 3 weeks with no pre-existing specification; achieved ~80% adoption within the first month, eliminating analytics-team dependency for ad-hoc reporting. • Built three-layer SQL validation (schema grounding, syntax guardrails, output plausibility checks), and integrated RESTful APIs with OAuth2 authentication to connect enterprise data services. • Designed and deployed a full production RAG architecture using Azure AI Search (hybrid dense + BM25 retrieval, semantic chunking, re-ranking, citation enforcement) with a five-layer hallucination-prevention framework (including answerability thresholds and confidence scoring), built and tuned against a 100-sample gold-set evaluation framework, reducing fabricated outputs to a near-zero rate within six weeks of human-in-the-loop feedback.

Machine Learning Engineer

Trans-Formation Logistics

Jan 2024 - Feb 2025 • 1 yr 1 mo

• Automated insights with an LLM pipeline, cutting decision-making time by 60% and boosting data accuracy by 35% for a state Department of Transportation (DOT) client. • Architected enterprise AI roadmaps for DOT modernization projects; designed Azure OpenAI-based solutions that reduced deployment complexity by 40% and established governance frameworks for responsible AI implementation. • Led cross-functional teams of up to 12 engineers to standardize reusable ML delivery workflows across data preparation, model validation, deployment, monitoring, and operational handoffs, supporting a $2M digital transformation initiative with 95% stakeholder satisfaction. • Designed scalable GenAI architectures with RAG pipelines for infrastructure analysis, implementing an AI Gateway for centralized model access control and monitoring across 3 production environments. • Designed and deployed a production-ready RAG system (Azure OpenAI, Azure AI Search, LlamaIndex) with evaluation benchmarks and A/B testing of retrieval strategies; packaged it into a repeatable consulting offering that directly supported new contract wins. • Integrated AWS Textract and custom neural networks to extract text from images and Word documents, speeding processing by 70% at 90% accuracy. • Fine-tuned and optimized open-source foundation models (including LLaMA and Mistral variants) with custom enterprise data, raising classification accuracy by 15% for automated feedback routing; evaluated fine-tuning approaches (LoRA, prompt-tuning) against prompt-engineering strategies on latency, accuracy, and cost trade-offs. • Implemented and optimized RAG systems integrating vector databases (Pinecone, FAISS) with custom embedding models to ground model responses and incorporate SME feedback. • Built segmentation and propensity models for a transportation client, enhancing revenue-assurance targeting; visualized results in Tableau and Qlik.

Data Scientist

Capital One

Jul 2022 - Jun 2023 • 11 mos

• Architected PySpark ETL pipelines on AWS EMR and Hadoop, processing billions of rows monthly and eliminating manual processing bottlenecks. • Ingested and unified disparate data using Python, SQL, and AWS & Salesforce APIs to deliver consolidated Snowflake data views for BI analytics. • Designed multi-layer ETL transformations in Snowflake via AWS EMR/Hadoop, boosting data retrieval performance by 40%. • Deployed containerized ML models using Docker and Kubernetes on AWS SageMaker with auto-scaling for 50M+ daily transactions, achieving 99.9% uptime. • Engineered PySpark data pipelines with Apache Airflow orchestration for anomaly detection, reducing pipeline failures by 75%; implemented Kafka for real-time streaming pipelines. • Built deep-learning models for customer-behavior prediction (TensorFlow 2.0) using attention mechanisms, improving accuracy by 28%. • Created an MLflow experiment-tracking system managing 200+ model versions, automating hyperparameter tuning and reducing model-selection time by 40%.