AI Coding — Friend Recommendation

Starter codebase exposes a `User` class with `friends` and `valid_recommend` helpers. Three sub-tasks build from logic-bug fix to AI-assisted random recommender to mutual-friends Top-K ranking. Requirements Q1 — Logic-bug fix in valid_recommend. Typical bug: the function does not exclude the user themselves and/or existing friends from candidates. Q2 — Implement random_recommend. Pick a random non-friend user. Prompt the AI with the class signature so the output matches the existing interface. Q3 — Top-K mutual-friends recommendation. Score each candidate by |friends(u) ∩ friends(candidate)|; return the top K. Walk the interviewer through time complexity and discuss how the metric would change for second-degree connections. Notes The starter User model and valid_recommend helper are pre-written — read them carefully before prompting the AI, otherwise it will fabricate a parallel API. The Q3 metric is the discussion point: be ready to argue mutual_friends vs jaccard_similarity vs graph_embedding_lookup, with rough complexity. Preparation Drill the LeetCode 1311 (Get Watched Videos by Your Friends) and 547 (Friend Circles) patterns to internalize friend-graph traversal. Rehearse a 3-sentence trade-off pitch for mutual-friend vs Jaccard vs embedding-based recommendation. Prep a clean prompt: "Given the User class with self.friends: set[User], implement random_recommend(user, k) that returns k random non-friend users; do not include the user themselves."