Sushma A

Ai Ml Engineer @Fidelity Investments

TX, US

Built end-to-end GenAI automation pipeline for Chief Investment Office, integrating portfolio optimization, market intelligence, and compliance, serving 500+ investment professionals. Developed predictive portfolio allocation models (XGBoost, Scikit-learn) using historical returns and macro data, improving risk-adjusted returns by 12% in live trading. Engineered real-time data pipelines (Apache Kafka, AWS Glue, Athena) for market feeds, news sentiment, and regulatory updates with 99.8% accuracy into structured event streams. Designed feature engineering framework generating time-series windows, cross-sector risk indicators, and microstructure metrics for ML and LLM-driven strategies. Implemented multi-agent orchestration (LangGraph, MCP, LLaMA 3 8B) with semantic routing, cutting manual memo/report creation time by 50%. Optimized finance-domain LLMs (FinGPT, Palmyra) using QLoRA + PEFT, reducing GPU memory usage by 60% while enhancing summarization and classification accuracy. Created prompt engineering system with few-shot learning, Chain-of-Thought, and dynamic context assembly for personalized market analysis and investment narratives. Built RAG pipeline (FAISS, Redis, Azure Vector DB, HNSW) combining BM25 + semantic reranking, boosting retrieval relevance by 35% for earnings and filings search. Deployed ML/LLM models via AWS SageMaker + Docker with CI/CD pipelines, cutting deployment cycles by 30% and ensuring version-controlled rollbacks. Established MLOps/LLMOps (MLflow) with automated retraining, drift detection, and LLM evaluation gates (BLEU, ROUGE, financial relevance) to maintain production-grade quality.

University of North Texas

Master's degree