Master 20 LeetCode patterns, not 300

A backend developer on X touched a nerve by saying interview prep is not about memorising 300 answers. The claim was that roughly 20 reusable patterns cover most coding interview questions people see on LeetCode-style screens. (x.com) That idea sounds small until you look at how interview prep is actually packaged. NeetCode’s roadmap groups practice into recurring buckets like arrays and hashing, two pointers, sliding window, heap or priority queue, graphs, and one-dimensional and two-dimensional dynamic programming instead of treating every problem as a brand-new species. (neetcode.io) A pattern is just a repeatable shape. If a problem says “longest substring” or “smallest subarray,” that often points to sliding window, which means moving one range across the input instead of restarting from scratch at every position. (leetcode.com) Two pointers is another one. When an array is sorted, you can start one pointer at the left end and one at the right end and close in like two people searching a shelf from opposite sides, which is why problems like Two Sum II and Container With Most Water fall into the same family. (leetcode.com) Heap, also called a priority queue, is the pattern for “keep the biggest” or “keep the smallest” without sorting everything every time. It works like a constantly updated shortlist, which is why it shows up in top-K and scheduling questions. (neetcode.io) Graph search is the pattern for anything that looks like places connected by roads. Breadth-first search moves level by level with a queue, while depth-first search pushes down one path first, and interview questions dress that up as islands, courses, flights, or social connections. (leetcode.com) Dynamic programming sounds scarier than it is. It means storing answers to smaller subproblems so you do not recompute the same work 50 times, like writing partial totals on sticky notes instead of solving the whole puzzle from zero on every move. (neetcode.io) This is why “do 300 random problems” often feels fake-productive. If you solve 10 sliding-window questions without learning the trigger words, the window rules, and the update pattern, you practiced trivia instead of learning the machine underneath it. (educative.io) The post landed because it matches how many candidates already study in 2025 and 2026. Guides across LeetCode, NeetCode, Intervu, DesignGurus, and other prep sites now organize practice around 10 to 20 recurring templates rather than giant unsorted lists, even when they disagree on the exact count. (leetcode.com) (neetcode.io) (intervu.dev) (designgurus.io) The catch is that pattern fluency is not the same as memorising a label. You still need to know the signal that picks the pattern, the time and space tradeoff it buys you, and the standard template you can write cleanly under a 35-minute interview clock. (intervu.dev) So “master 20 patterns” is not a shortcut around practice. It is a way to compress practice, because 150 well-chosen problems mapped to recurring structures can teach more than 500 disconnected ones learned as isolated tricks. (neetcode.io) (educative.io)