• as a leader of the ML strategy unit consisting of ML product strategists, ML data strategists, labeling managers, and labeler educators, clarified each position’s R&R and developed the overall work pipeline.
• enhanced fine-tuning dataset quality for the chatbot, achieving a milestone of 2 million users along with 18% of retention rate.
• made a new persona chatbot from a to z, reaching 50,000 DAUs in CIC project.
• conducted in-depth research on the diverse LLM FT techniques and designed training data schema for SFT, Instruction FT and RLHF through intensive discussion with ML researchers.
• proposed a ‘regeneration’ feature of a chatbot by implementing qualitative and quantitative data segment analysis using Python, generating data worth ₩5.5M every month.
• organized and trained a labeling team consisting of 2 labeling managers, 5 educators and 40+ freelancers for labeling tasks.
• optimized the pipeline from data construction to quality check system, which reducted labeling time by 45%.
더보기