You will need data on relevant results for a broad swath of queries to fine tune your ranking models against, and will need to spend millions on infrastructure just to crawl and store an index of the entire web.