• Worked on AI-driven fraud detection systems for HSBC corporate and card transactions, processing millions daily, reducing fraud exposure by 35%, accelerating risk response by 40%, ensuring SOX and SOC2 compliance, and enabling agentic AI interventions.

• Built robust end-to-end data pipelines using Python, SQLAlchemy, and Pydantic, storing structured data in PostgreSQL with 99.5% schema validation accuracy, improving audit readiness by 30%, and supporting real-time monitoring through AWS Bedrock integration.

• Engineered transaction-level NLP features, including risk triggers, anomaly descriptors, and temporal sequences using spaCy and rule-based

matchers, improving feature quality by 25%, enhancing fraud classification precision by 18%, and integrating agentic AI feedback loops.

• Fine-tuned Longformer and FinBERT models on 200K+ labeled HSBC transactions, improving F1 scores from 72% to 88%, reducing compute costs by 40% with LoRA and PEFT, maintaining strong security and compliance controls.

• Designed real-time Retrieval-Augmented Generation (RAG) pipelines using LangChain, FAISS, AWS Lambda, and dashboards, increasing historical

query accuracy by 30%, reducing investigation time by 35%, and providing fully auditable retraining workflows.

• Leveraged AWS Bedrock and agentic AI for automated anomaly detection, real-time transaction risk scoring, and dynamic decision-making, enabling HSBC to proactively mitigate fraud, optimize operational efficiency, and maintain enterprise-grade compliance.

• Developed a financial analytics platform analyzing market data, trade settlements, and portfolio positions, generating risk scores and investment

recommendations, reducing decision-making latency by 20% and enhancing operational efficiency across multi-asset global portfolios.

• Built scalable batch and streaming pipelines using Azure Data Factory, Databricks, and PySpark, processing structured and unstructured data from Snowflake and PostgreSQL, enabling real-time dashboards, regulatory reporting, and actionable insights for portfolio managers and analysts.

• Applied NLP and transformer models to trade notes, research reports, and earnings call transcripts, predicting market risks, trade anomalies, and

compliance issues, increasing prediction accuracy by 17% and supporting proactive risk mitigation strategies across teams.

• Developed predictive models using gradient boosting, LSTM time-series, and contextual embeddings for asset performance and market trends,

deploying via Azure ML and FastAPI microservices in collaboration with DevOps and cloud engineering teams.

• Optimized model performance with Optuna hyperparameter tuning and SHAP interpretability, achieving RMSE 5.9, Precision@3 80%, and AUC 0.89, outperforming legacy financial scoring models while ensuring regulatory compliance and audit-ready transparency.

• Implemented real-time trade monitoring, alerting, and exception handling services using Kafka, Docker, and Kubernetes, working closely with

DevOps and platform teams to maintain 99.7% uptime and streamline operational risk management.

• Automated model retraining, validation, and drift detection workflows using Azure ML pipelines, MLflow, Prometheus, and Grafana dashboards,

monitoring predictive performance and operational KPIs, achieving 93% SLA compliance and data-driven investment outcomes.