Built a LangGraph agent that runs parallel Bright Data searches (Google, Bing, Reddit) with GPT-4 synthesis, reducing time- to-insight by 5× (80% faster). • Improved factual precision by 40% using Pydantic-validated outputs and layered prompt engineering. • Implemented async snapshot scraping with Bright Data—retrieved thousands of Reddit posts per query and reduced failed-crawl rate to <5%. • Reduced LLM cost per query by 60% through model tiering and batching, while preserving end-to-end accuracy. Github