Senior AI/ML Engineer at Shift Technology

Delivered production AI capabilities for Shift Claims Fraud Detection and Shift Underwriting Risk Detection using Microsoft AI Foundry. Built document intelligence pipelines to process claim forms, policies, invoices, medical records, and underwriting documents into structured evidence. Designed supervisor-agent workflows for document analysis, entity resolution, policy retrieval, risk explanation, and case summarization. Integrated Azure AI Search, graph services, model endpoints, and insurer APIs with production-grade agent execution, retries, shared state management, and approval workflows. Implemented enterprise hybrid RAG with semantic retrieval, reranking, metadata filtering, citations/provenance, confidence scoring, contradiction detection, abstention logic, and retrieval evaluation datasets. Combined fraud prediction models, anomaly detection, entity resolution, graph analytics, and LLM narratives while preserving model scores, rule triggers, evidence sources, and reason codes.

Senior AI Consultant & Data Engineer at FTI Consulting

Delivered a cloud-based financial crime intelligence platform supporting KYC, entity resolution, transaction monitoring, risk scoring, policy research, and regulated case investigations. Built scalable GCP data pipelines with Cloud Storage, BigQuery, Dataproc/Spark, Pub/Sub, Dataflow, and Document AI for ingestion, transformation, validation, and enrichment of compliance data. Standardized customer/account/transaction/document schemas and improved data availability time from hours to ~15 minutes. Developed a grounded Gemini and Vertex AI assistant using retrieval, metadata filtering, citations, structured summaries, and source-level access controls. Designed Vertex AI Agent Builder workflows with coordinator and specialist agents for document review, policy verification, entity analysis, risk explanation, and case summarization. Connected agents with BigQuery, Document AI, and graph services using function calling, workflow state management, approval gates, and reviewer routing. Produc

Senior Machine Learning Engineer at Stripe

Enhanced Stripe Radar 2.0, a real-time payment risk platform combining ML models, deterministic rules, merchant configurations, and manual review workflows. Engineered point-in-time ML features from 500M–1B daily events across payments, merchants, devices, disputes, chargebacks, and accounts using AWS S3, Glue, EMR/Spark, Python/Scala, and SQL. Built fraud detection models using XGBoost and SageMaker with delayed-label handling, class weighting, time-based validation, calibration, and threshold optimization. Improved fraud capture by ~12% while reducing false-positive payment blocks by ~10% under fixed review capacity. Designed low-latency online inference architecture with DynamoDB, SageMaker endpoints, rules engines, and merchant configurations achieving ~40ms p99 decision latency. Developed Kinesis-based event pipelines and CloudWatch monitoring for disputes/chargebacks/reviews, model drift, temporal replay, and safe model rollback. Applied graph analytics and BERT-based NLP for p

Machine Learning Engineer at Kira Systems

Developed ML capabilities for the Kira Contract Analysis Platform, including Quick Study and Smart Fields, to extract, review, and analyze contract provisions. Built document processing pipelines for PDF/DOCX/scanned documents and OCR while preserving page/paragraph/text-span mappings for source-level highlighting. Trained clause classification and information extraction models using SVM, gradient boosting, CRF, and LSTM architectures with class weighting, hard-negative mining, and threshold tuning. Improved clause classification F1 from ~0.82 to ~0.90 through optimization and error analysis. Designed active learning, batch inference, and regression testing workflows for uncertain samples and custom Smart Field versions. Processed ~150K pages monthly, reducing annotation effort ~35% and improving document processing speed ~60%. Collaborated with engineering/product teams through Agile/Scrum cycles, validation, and production releases.