LLM Training and span annotation
This project focused on factuality evaluation and annotation of AI-generated text for fine-tuning conversational AI models. Tasks involved span-by-span analysis of text to verify factual accuracy, identifying misleading claims, and categorizing spans into fine-grained labels based on factuality rubrics. Additionally, side-by-side comparisons were conducted to assess response quality, focusing on factual alignment, linguistic fluency, and instruction-following accuracy. The project encompassed text classification, relationship analysis, and prompt refinement for improving AI outputs. Adhering to strict quality measures, I ensured accuracy by conducting detailed research and documenting findings meticulously. This work contributed significantly to the fine-tuning of advanced LLMs for safe and reliable user interactions.