Sason Torosean

About

ML/NLP practitioner and technical founder who built an AUP risk-scoring model at Bill.com using TF-IDF + SVD features with XGBoost, and previously created automated data-quality guardrails for a Global Equity Risk stacked ML model at Thomson Reuters. Recently built a RAG-based chatbot for PaymentJock’s Home Affordability Probability product using embeddings and a local vector database (FAISS/Chroma), improving answer quality through chunking rather than expensive fine-tuning.

Experience

Education

Key Strengths

• Applied NLP feature engineering (TF-IDF n-grams + SVD) to improve risk scoring with XGBoost
• Built robust cross-validation pipelines to tune NLP feature parameters (n-grams, SVD components)
• Designed and implemented automated data quality/QA test suites to prevent dirty data from impacting model training
• Cross-functional collaboration with Quant, Product, and QA to maintain reliable ML pipelines
• Built and evaluated a RAG chatbot using embeddings + local vector DB; improved coherence via chunking instead of costly fine-tuning
• Production-minded Python workflow design using reusable OOP components, parallelization, Docker, and memory management
• Strong stakeholder alignment and communication to move prototypes into production; resolved prior delivery issues by improving communication