The real challenge in search is not how to pick the best search engine framework or how to find a match, but how to bring the most relevant results first. Modern search engines have been heavily using machine learning to understand what is relevant for their users, but this approach requires tons of time and effort. In this talk, we will show how to implement from scratch a clever search for your application using just a little bit of planning and standard features provided by any search engine framework: boosting, analyzers, tokenizers, fuzziness, facets, penalties, etc. We will also review some core concepts and show how to fix common relevancy problems.