Friend Recommendations: Solution Walkthrough
Medium Priority15 min readUpdated November 11, 2026
Friend Recommendations is the library behind a small social app. It tracks friendships, blocked users, and deactivated accounts. Each user is an integer id, and the social graph is loaded from a text file of changes applied in order.
The round has four phases:
- Phase 0: Understand the code. Read the existing implementation and fill in three blanks.
- Phase 1: Fix one-sided friendships. A bug report says a friendship shows on one user's profile but not the other's.
- Phase 2: Suggest friends. Add
recommend(graph, user, k=5), which suggests people by mutual friend count. - Phase 3: Make suggestions faster. Put recommendations behind a cache that stays correct while the graph changes, at 10,000 users and up to 1,000,000 requests (300,000 in Python).
The code is small, and the Phase 2 algorithm is the familiar friends-of-friends count. What trips people up is the product rules. Blocks apply in both directions, deactivated accounts vanish, and a mutual friend only counts if both people can see them.
Phase 3 then turns on one question you have to answer exactly: when a single edge changes, whose recommendations can change with it?