Fahad Yousaf

• Identified and resolved client-level threshold inconsistencies by analyzing feature distributions and fraud patterns, designing client specific scoring thresholds to prevent over or under scoring across programs with similar transaction volumes and fraud behavior. • Debugged production model underperformance by tracing score degradation back to missing or incorrectly generated statistical features, and coordinated with engineering teams to restore correct stat generation in production pipelines. • Handled feature engineering pipelines using SQL based statistical aggregation and Spark based Python pipelines, transforming raw transaction data into production-ready fraud features at scale. • Monitored production fraud systems using Grafana dashboards and Airflow DAGs, ensuring system reliability and timely detection of pipeline or scoring anomalies.

• Enhanced an internal Retrieval-Augmented Generation (RAG) chatbot, improving response quality by 50% using hybrid search (semantic + keyword-based retrieval). • Achieved 25% reduction in fraudulent activity through detailed analysis of card programs, merchants, and cardholders. Implemented customized fraud prevention model, reducing fraudulent transactions by 30%. Collaborated with 48 clients to enhance security measures and protect against fraud. • Reduced retrieval latency by 50% through an optimized indexing method. • Worked on a fraud detection engine, optimizing fraud detection on 4-5 million daily transactions, leading to 40% fraud transaction catch rate and 21% fraud value catch. • Improved Merchant Scoring Model, increasing performance by 2% in volume catch and 5% in value catch while maintaining near-zero genuine declines. • Functionalized and optimized an ML pipeline in Jupyter Notebook, reducing manual configuration by 20% and streamlining model training and scorecard creation. • Set up and debugged a Redis cluster using Docker, doubling search speed and improving system scalability.

• Developed a RAG-based Chatbot enabling users to query PDFs, reducing resolution time by 30%. • Utilized LangChain for LLM chains, Qdrant as the vector database, FastAPI for the backend, and Streamlit for the frontend. • Implemented a custom address extraction algorithm using SpaCy (NER), reducing manual effort by 50%. • Designed a face extraction algorithm using OpenCV for processing images and PDFs. • Designed and optimized Azure Data Factory ETL pipelines, enhancing data transformation and integration. • Integrated automated failure alerts in ETL pipelines, reducing issue resolution time by 40% • Optimized SQL queries and ETL pipelines, achieving a 25% performance boost. • Built an ETL system syncing 1M+ daily ledger entries from QuickBooks to NetSuite, cutting manual effort by 60%.

Successful Project Completion in Muscular Dystrophy Rehabilitation (FYP): -- Completed a project aimed at rehabilitating patients with muscular dystrophy, aiding doctors in understanding muscular impairments. -- Achieved a 100% on-time project completion rate, showcasing commitment and project management skills. -- Trained a random forest model on a raw dataset collected from Punjab Institute of Neuro Science (PINS) General Hospital and Jinnah Hospital, Lahore, achieving an accuracy rate of 73%, surpassing the targeted 60%.