AI "Pet Voice" Generator (Mini-Program Prototype) — prompt engineering and LLM output evaluation
Performed end-to-end LLM output evaluation and quality judgment for generated Chinese text by testing prompt strategies and refining instructions for tone, accuracy, and reasoning. Assessed persona consistency and output reliability using controlled generation prompts with an LLM API workflow. • Evaluated tone control and reduced errors such as hallucinations and weak reasoning • Compared multiple prompt variants to select the best instruction set • Applied criteria-based checking for factuality, clarity, and coherence • Iterated prompts to improve response quality and persona adherence