AI/ML Engineer at Fidelity Investments

Developed an AI-driven portfolio intelligence assistant using FinBERT and DeBERTa, integrating retirement data, brokerage activity, and client interactions; improved advisor decision efficiency by 22% and early risk-flag prediction accuracy by 17%. Engineered scalable, privacy-preserving ETL pipelines using Airflow, Spark, AWS Glue, and Delta Lake to compute investor risk scores, trading-behavior metrics, and client embeddings, reducing data processing latency by 35% and improving compliance audit coverage by 20%. Built client-behavior embeddings by combining FinBERT outputs with graph-based relationship signals, finding latent investment patterns, reducing high-risk escalations by 25%, and increasing portfolio recommendation relevance by 18%, achieving F1 0.87, Recall 0.90, and AUC-ROC 0.92. Fine-tuned FinBERT and DeBERTa models with adapters and LoRA for anomaly and fraud detection in trading accounts, increasing early fraud detection rates by 28% and improving NDCG@10 from 0.62 to 0