Muskan Sahetai

• Developed AI-driven fraud detection and transaction monitoring models processing millions of daily banking transactions, reducing fraud by 30% and improving real-time risk decision latency by 40%, enhancing overall financial security and operational efficiency. • Built scalable Python and SQL-based pipelines normalizing multi-source banking data into PostgreSQL and internal feature stores with 99.7% schema validation accuracy, enabling reliable internal APIs, analytics platforms, and downstream AI models for enterprise financial operations. • Engineered NLP-based transaction enrichment and categorization models using spaCy and FinBERT to identify merchant details, recurring patterns, and anomalies, increasing classification precision by 18%, supporting enhanced credit underwriting and automated risk assessment workflows. • Fine-tuned transformer models (Longformer, FinBERT) on 250K+ labeled transactions, boosting fraud detection F1-score from 74% to 89% while reducing inference costs by 35% through LoRA and PEFT optimization for large-scale production deployment. • Designed real-time RAG and agentic AI workflows leveraging LangChain, FAISS, and AWS Bedrock, improving investigation accuracy by 28%, reducing manual review time by 35%, while maintaining SOC2 and PCI-DSS compliant, high-availability ML services.

• Developed an Intelligent Financial Insights Platform to process and analyze large-scale market and client interaction data, improving analytical throughput by 22% and enabling research teams to make faster, data-driven decisions. • Built scalable ETL pipelines using Azure Databricks, Delta Lake, and Spark Structured Streaming, integrating structured financial datasets with unstructured analyst notes, enabling faster model training and actionable insights for risk, research, and analytics teams. • Designed and implemented ML models using LightGBM, Transformer-based Autoencoders, and scikit-learn to classify financial instruments, detect market anomalies, and forecast trends, achieving 89% F1-score in classification and 94% recall in anomaly detection. • Applied RoBERTa and Sentence-BERT embeddings with logistic regression to extract insights from analyst reports and transaction logs, reducing manual review workload by 21% and improving operational efficiency for research teams. • Optimized LightGBM and DeBERTa-v3 models using Optuna for Bayesian hyperparameter tuning, and applied SHAP and LIME explainability techniques to interpret predictions, enhancing stakeholder trust in AI-driven insights. • Containerized ML models with Docker and deployed on Azure Kubernetes Service with KServe, exposing REST and gRPC APIs integrated into internal dashboards for real-time analytics and faster operational decision-making. • Implemented continuous training pipelines using MLflow and Kubeflow, monitored model drift with Evidently AI and Grafana, maintaining model accuracy above 91% and ensuring reliable, scalable, data-driven operations across S&P Global’s financial research systems.

With a stellar CGPA of 9.37, I graduated with a bachelor's degree in May 2024, accompanied by the publication of two research papers showcasing my dedication to academic excellence. Complementing my theoretical knowledge, I completed four internships, gaining hands-on experience and a nuanced understanding of industry practices. As a Computer Society of India (CSI) member, I participated in group activities, discussions, and collaborative projects, fostering my passion for technology and expanding my professional network. Now, equipped with a robust academic foundation, practical skills, and a commitment to lifelong learning, I am poised to contribute meaningfully to the technology sector.