Kevin Benassuly

– Built a rubric-grounded automated grading pipeline for large universities, using ML for high-confidence cases and LLMs for ambiguous long-tail grading, reducing manual grading by 60%. – Implemented explainability-gated predictions and automated detection of grader bias, inconsistencies, and rubric issues to continuously improve grading quality.

– Managed temporarily a cross-functional team of data scientists, data analysts and data annotators to automate IT support ticket classification (50+ classes). – Designed and built the incident categorization algorithm that makes IT management more efficient, resulting in a 20% reduction in ticket resolution time. – Researched LLM applications for virtual assistant and text summarization projects aimed to relieve IT experts from time-consuming and repetitive queries, accounting for 30% of their tasks.

– Trained a Named Entity Recognizer to extract meaningful financial insights from corporate annual reports. – Built a fast-paced semantic search algorithm using BERT architecture and Elasticsearch indexing, resulting in a 15% reduction in time complexity. – Implemented various time series forecasting techniques to predict key company performance measures as a new product feature, increasing conversions by 15%.