150+ data-entry specialists manually visited supplier websites, copying product specifications into standardized Excel sheets — slow, repetitive, error-prone, and scalable only by hiring more people.
Led the engineering team that built an LLM-powered extraction platform: automated crawling (Crawl4AI, Playwright), LLM structured extraction, FastAPI services, Redis queues, Dockerized microservices, and a Next.js control panel — turning entire product categories into clean, schema-consistent Excel automatically. I knew this workflow down to the keystroke: many of the operators trained at my own institute.
~95% of the manual workload automated across the client's four target supplier sites, at ~95% extraction accuracy. The data team now verifies instead of retypes — and the client is expanding the system into a general-purpose extraction platform.







