AI微调工程师 – Data Labeling & AI Training Lead
TextPrompt Response Writing SFT
Led the end-to-end design and construction of high-quality SFT and RLHF data sets for financial-sector LLM fine-tuning. Oversaw data cleaning, diversity sampling, and adversarial prompt injection to ensure data quality. Defined and implemented manual evaluation protocols for model outputs, including preference and blind tests.• Created and reviewed over 120,000 instruction-response text pairs for supervised fine-tuning (SFT)
• Designed multi-stage QA and DPO alignment experiments for labeled datasets
• Led cross-team efforts on annotation protocol and labeling platform standards
• Monitored annotation feedback cycles to close quality gaps in dataset creation
Led the end-to-end design and construction of high-quality SFT and RLHF data sets for financial-sector LLM fine-tuning. Oversaw data cleaning, diversity sampling, and adversarial prompt injection to ensure data quality. Defined and implemented manual evaluation protocols for model outputs, including preference and blind tests.• Created and reviewed over 120,000 instruction-response text pairs for supervised fine-tuning (SFT) • Designed multi-stage QA and DPO alignment experiments for labeled datasets • Led cross-team efforts on annotation protocol and labeling platform standards • Monitored annotation feedback cycles to close quality gaps in dataset creation
2022 - 2024