LLM Article Summarizer

Built an LLM-powered article analysis tool that takes arbitrary web articles as input and generates structured summaries using large language models. The core focus was exploring how prompt design — instruction framing, few-shot examples, output format constraints — influences the quality, consistency, and variability of model outputs.

The project demonstrates concrete differences between naive and engineered prompts: structured prompts with explicit role instructions and output templates consistently produced more accurate and stable summaries compared to open-ended prompts.