Avtar Singh Panwar
About
Detail
New Jersey, United States
Contact Avtar regarding:
groups
Networking
Résumé
Jobs
verified_user
0% verified
Data ScientistSardineJul 2025 - Current (1 year 2 months)• Developed graph-based fraud detection models on real-time payment networks using PySpark and XGBoost, identifying mule networks and increasing fraud capture rate by 12% globally. • Engineered streaming feature pipelines using Kafka and Spark Structured Streaming, reducing fraud detection latency from 800ms to 250ms across high-volume, real-time transaction processing systems. • Collaborated with compliance and risk stakeholders to optimize transaction monitoring thresholds, enabling scalable onboarding of financial institutions while ensuring regulatory compliance and minimizing false positive alerts. • Implemented continuous model monitoring and automated retraining pipelines using MLflow and Airflow, maintaining stable fraud detection p- Data ScientistHCL TECHJun 2021 - Dec 2023 (2 years 7 months)• Designed gradient boosting fraud models on card and wallet transactions, increasing detection precision by 13% while keeping false alerts controlled during peak transaction spikes. • Built probability of default and loss given default models using transactional and behavioral signals, improving risk differentiation accuracy by 9% across unsecured digital lending portfolios. • Implemented end-to-end feature engineering pipelines with automated validation and versioning, reducing model deployment turnaround time from four weeks to ten days across client programs. • Applied SHAP and stability metrics to explain model decisions, cutting regulatory review cycles by three weeks and enabling faster approvals from compliance and audit stakeholder
- Data AnalystTalaApr 2019 - May 2021 (2 years 2 months)• Analyzed mobile lending application data to identify credit risk patterns, building Power BI dashboards that reduced early default rates by 6% across 1.2M loan applications in India market. • Partnered with product and risk teams to define underwriting metrics, validating model inputs through SQL analysis that improved approval accuracy by 4% within six months for borrowers. • Designed end-to-end ETL workflows for transaction and repayment datasets, automating data quality checks that cut reporting latency from five days to one day during monthly risk reviews. • Conducted A/B analysis on loan pricing experiments, quantifying customer response and revealing segments that increased net interest revenue by 8% without elevating delinquency ac
Education
verified_user
0% verified
- Master of Science in Business Analytics & Data ScienceNew Jersey City UniversityJan 2024 - Dec 2025 (2 years)
- Bachelor of Science in Computer ScienceVikram universityMay 2016 - May 2019 (3 years 1 month)