Project Manager — Data annotation, prompt engineering, and dataset quality control for multimodal LLM fine-tuning (2010-present)
Text
Fine-tuningLed multimodal dataset construction for localized AI customer service and automated O&M model training by driving rigorous annotation and quality assurance processes. Implemented prompt-engineering and instruction-oriented data work to support LLM fine-tuning with business texts, voice dialogues, and field images as training inputs. Increased annotation accuracy to over 98% through multi-round cross-verification and dataset quality control. • Defined and enforced Annotation Guidelines for consistent labeling outcomes. • Orchestrated multimodal annotation workflows covering business text and voice dialogue content. • Built dataset QA and cleaning pipelines that combined manual labeling with machine pre-labeling. • Enabled downstream LLM fine-tuning by delivering high-quality training corpora to the model optimization pipeline.