Security Engineer – Machine Learning & Data Science
Paysecure Technologies Pvt. Ltd. · Aug 2025 – Present
- Engineered and trained a multi-class XGBoost fraud detection model end-to-end - feature engineering, statistical validation, hyperparameter tuning on 2.5M+ paid vs. 25K chargebacks, achieving 86% test accuracy (0.829 macro F1) and 93.5% recall / 95.5% precision on the highest-risk tier.
- Extending the platform with Generative AI: LLMs + RAG over historical fraud cases, embeddings/vector search, and LangChain/LangGraph agents for evidence retrieval, suspicious-entity investigation, and grounded risk summaries.
- Built an AI-powered Slack bot (Python, Claude API, MySQL, MongoDB, GitHub API) that auto-generates and executes JUnit test suites from live GitHub diffs across a 3,400+ file Java payment-gateway codebase, automating manual commit-review testing.
- Designed a two-stage behavioral Trust Score system (hierarchical K-Means → 5 risk tiers → downstream XGBoost classifier, 0.909 macro F1, 0.996 AUC-OvR) - translating model output into business impact: auto-declines cut from 36% to 3.4%, auto-approvals up from 14% to 78%.
- Used Power BI and statistical EDA to derive features and drive down false-positive rates, built distributed pipelines scaling OSINT ingest throughput 9.7x (6K → 58K records/sec) across 5.5TB, and shipped a FastAPI layer sustaining sub-100ms inference.