Prompt Writer
Designed and developed complex, synthetic datasets to facilitate the Supervised Fine-Tuning (SFT) of Large Language Models. The project scope focused on creating sophisticated, multi-turn conversational architectures that challenged model reasoning, instruction-following, and contextual memory. Specific Data Labeling Tasks Performed: Authored end-to-end multi-turn dialogues from scratch, simulating complex user-AI interactions. Developed "edge-case" prompts requiring the model to navigate conflicting instructions or maintain a specific persona over long-form conversations. Generated high-quality ground-truth responses to train models in specific reasoning tasks and logical deduction. Quality Measures Adhered To: Followed rigorous rubrics for linguistic precision, factual accuracy, and logical consistency. Adhered to strict alignment guidelines regarding model safety, tone, and helpfulness. Participated in iterative feedback cycles to ensure data met high-complexity benchmarks for model training. 2. Outlier (Generalist Data Annotation) Project Description: Contributed to the optimization of multimodal AI models through high-fidelity data refinement and Computer Vision (CV) tasks. The project scope involved improving model perception and object-handling capabilities across various media formats, including video, image, and audio. Specific Data Labeling Tasks Performed: Executed complex video manipulation tasks, including removing specific items and using text prompts to direct the model in re-inserting objects (inpainting). Performed entity tagging and semantic labeling of products, people, and landmarks to enhance object detection datasets. Transcribed audio and video content to provide ground-truth data for speech-to-text and natural language processing models. Quality Measures Adhered To: Maintained 95%+ accuracy ratings by strictly adhering to granular, project-specific style guides. Ensured temporal consistency in video tasks to minimize artifacts and training noise. Collaborated within a high-volume environment while meeting strict deadlines and inter-annotator agreement standards.