Aishwarya Agrawal

About Aishwarya Agrawal

Thank you for visiting my profile!Currently, I am working as a Senior Associate Data Scientist at BNY Mellon, where I’ve been working in the anomaly detection and fraud analytics space across various payment types. I focus on designing and deploying machine learning models that are both scalable and resilient, using techniques such as time series modeling, unsupervised learning, and statistical pattern recognition to detect outliers and minimize risk across payment rails like Wires, ACH. I completed my Master’s in Artificial Intelligence and Innovation at Carnegie Mellon University (CMU) in Pittsburgh. During my capstone with BNY, I worked on building a Retrieval-Augmented Generation (RAG) based question-answering system, combining knowledge graphs, AI agents, and large language models (LLMs). We evaluated the system using RAGAS, focusing on LLM reliability, prompt engineering, and structured reasoning.I’m highly inspired to explore new ways AI and data science can be applied in meaningful, real-world settings — from financial systems to enterprise tools. I’m particularly passionate about responsible AI, guardrails for generative models, explainability, and building ethical, transparent AI systems that organizations can trust.My technical skills include Python, PyTorch, SQL, Data Modeling and Data Science, and I enjoy working on projects that combine natural language processing (NLP), LLM evaluation, and model interpretability.I’m always curious, always learning — and always looking to collaborate on impactful, forward-thinking AI projects.Let\'s connect: a••••••••@gmail.com

Experience

1. Senior Associate Data Scientist

   BNY

   Jul 2024 — Present · Pittsburgh, PA, US

   Anomaly detection solution for high-value (>$1M) commercial wire transfers• Complete ownership - developed and productionized real-time anomaly detection pipeline, used by 4 lines of business• Achieved a 92% recall and 51% precision evaluated on daily 24K wire transactions• Optimized end-to-end latency from 50 mins to 45 sec by implementing feature computation in SnowparkLLM Classification Agent for payment slicing• Built Autogen-based GPT-4o agent to classify SWIFT messages by payment type (debt vs. non-debt)• Deployed to production, boosted downstream anomaly-detection precision by 6% by more reliable message classification

Education

• The Bhavans' Prominent School

  SSC

  2015

• Institute of Engineering & Technology DAVV, Indore

  Bachelor of Engineering - BE, Computer Science

  2017 — 2021

• The Bhavans' Prominent School

  HSC

  2017

• Carnegie Mellon University

  Master's degree, Artificial Intelligence and Innovation