Quasar-V4 Persona Training Data Annotation
In the Quasar-V4 project, I was responsible for both the data annotation and the design of the model’s reasoning persona. I created structured datasets that established how the model should process information, break down problems, and communicate in a consistent voice. The persona design centered on giving the model a self-reflective reasoning style: it restated inputs, recalled relevant knowledge, structured subproblems, reasoned step by step, verified its own work, and then presented polished answers. Through annotation, I encoded not only factual correctness but also tone, adaptability, and conversational flow, ensuring the model could shift seamlessly across domains such as math, coding, and natural dialogue. This approach transformed raw text data into a training signal that captured both reasoning process and personality traits, enabling Quasar-V4 to align more closely with human-like structured thought and clear communication.