Masoud Rezvani

I am Masoud Rezvani (Masoud Rezvaninejad), an ML Engineer with 5+ years in data, ~3 of them deploying and operating production ML systems. MSc in Data Science from the University of Amsterdam. Currently an AI Engineer Intern at Talk360; previously fraud-detection and pricing models at Baly.iq and Snapp! scoring ~200K trips/day (~$500K/year prevented losses).

What I’m building:

• Learning-to-Rank News Recommender — hybrid BM25 + FAISS candidate generation into a LambdaMART ranker with MMR diversity; position-bias correction via IPW (+6% NDCG@10).
• LLM Support-Ticket Triage — Llama-3.2-3B fine-tuned with 4-bit QLoRA for structured-JSON classification (99% eval token accuracy, 0.75% of params trained).
• Production RAG at Talk360 — hybrid retrieval + cross-encoder reranking, 98% faithfulness, prompt-injection defenses.

Published in Expert Systems with Applications (2025): WDAE-GAN for credit-card fraud detection (DOI).