Senior AI Engineer at Con Edison

Built production agentic AI on the Model Context Protocol (MCP) for field and operations teams, enabling multi-step operational tasks via tool calls to governed internal systems. Designed MCP servers exposing equipment records, manuals, safety procedures, and work management APIs, and implemented MCP clients integrated with LangGraph orchestration (planning, tool selection, retries, state management). Delivered retrieval augmented generation with hybrid retrieval, reranking, and strict citation enforcement. Engineered serving on Azure using Azure OpenAI, Azure AI Search, and AKS for low-latency/high-throughput workloads. Implemented layered guardrails including prompt injection defense, tool permissioning/scoping, sensitive data handling, and content filtering. Established evaluation pipelines for agent trajectories (tool-use correctness, task completion, faithfulness, and safety) and added access control/audit logging for all MCP tool calls. Also built and deployed ML models for outag

Senior AI and Machine Learning Engineer at Oscar Health

Built real-time agent and retrieval assistants for member and clinical support by calling internal tools/APIs for benefits, eligibility, and prior authorization status, returning grounded, cited answers within seconds. Implemented retrieval augmented generation grounded in member plan and clinical documentation using embeddings, vector search, and reranking. Developed standardized tool/function calling interfaces for model access to internal data sources. Fine-tuned and evaluated large language models for summarization, classification, and extraction with guardrail layers for grounding, safety, clinical appropriateness, and prompt-injection defense as a release gate. Built real-time prior authorization triage/risk/cost models deployed on SageMaker, and deployed generative AI services on Bedrock, SageMaker, and Lambda. Established MLOps foundations (model registry, CI/CD for model and prompt changes, automated testing), caching/routing across models by task, instrumentation/observabilit

Senior Machine Learning Engineer at U.S. Bancorp

Built and deployed real-time fraud scoring services on Azure, producing decisions within milliseconds for transaction authorization flows. Developed credit risk/propensity models with both batch and real-time endpoints, and built NLP models for servicing notes/communications as inference services. Established MLOps using MLflow and Azure ML (tracking, registry, deployment) and engineered low-latency containerized inference on AKS. Created feature pipelines on Databricks/Spark, implemented CI/CD for training/deployment with testing/versioning/rollback, and set up monitoring for drift/performance/latency with automated alerts/retraining. Partnered with model risk/compliance for governance and mentored engineers on MLOps and reliable serving.

Senior Data Scientist at GEICO

Developed claims intake triage scoring to predict severity and complexity and route claims to the correct adjuster queue within seconds. Built document NLP to extract key fields from claim documents and adjuster notes, reducing manual data entry. Implemented fraud/anomaly detection for unusual claims and supported special investigations. Built pricing and risk models using policy/driver/telematics data. Engineered scalable features using Spark on AWS EMR, trained/deployed/versioned models on SageMaker, and integrated ingestion/transformation pipelines using S3 and Glue. Performed backtesting and champion-challenger comparisons, and implemented monitoring and scheduled retraining to keep performance stable under changing claim mix and seasonality with internal governance documentation.