AI Engineer Problems › API Design and Debugging

Bug Hunt: Movie Recommendations API

Medium · Code Comprehension · API Design and Debugging · Asked at Amazon

Bug Hunt: Movie Recommendations API

You are reviewing a Django REST Framework codebase for a Movie Recommendations API built by a junior developer. The API lets authenticated users rate movies, manage a watchlist, and get personalized recommendations based on their activity.

The test suite has 6 tests, and they are all failing. Your job is to find and fix the 6 bugs hidden in views.py so that every test passes.

You should only edit views.py. The models, serializers, URL config, and tests are all correct.


API Reference

Ratings

Method Endpoint Description
POST /api/movies/<id>/rate/ Rate a movie (1-10). Creates a new rating (201 Created) or updates an existing one (200 OK).
GET /api/movies/<id>/my-rating/ Get the current user's rating for a movie.
GET /api/movies/<id>/average/ Get the average user rating (public, no auth required).

Watchlist

Method Endpoint Description
POST /api/movies/<id>/watchlist/ Toggle a movie on/off the user's watchlist.

Recommendations

Method Endpoint Description
GET /api/movies/recommendations/ Get up to 10 personalized recommendations.

Models

Movie
  - id (auto)
  - title (string)
  - year (integer)
  - genre (string, comma-separated e.g. "Crime,Drama")
  - rating (float, 0-10, average critic rating)
  - description (text)
  - popularity (integer)

UserRating
  - id (auto)
  - user (FK -> User)
  - movie (FK -> Movie)
  - score (integer, 1-10, the user's personal rating)
  - created_at (auto timestamp)

Watchlist
  - id (auto)
  - user (FK -> User)
  - movie (FK -> Movie)
  - added_at (auto timestamp)

Recommendation Algorithm

The get_recommendations endpoint uses this logic:

  1. No activity: If the user has zero ratings AND zero watchlist items, return a getting-started message.
  2. Rated movies:
    • Rated HIGH (score > 5): Find similar-genre candidates with rating >= 7.0, excluding the source movie. Apply a 1.2x multiplier.
    • Rated LOW (score <= 5): Find different-genre candidates with rating >= 7.0, excluding the source. No multiplier.
  3. Watchlisted movies (skip if also rated):
    • Find similar-genre candidates with rating >= 7.0, excluding the source. No multiplier.
  4. Scoring:
    • genreScore = matching_genres / source_genres
    • ratingScore = candidate_rating / 10
    • Rated-high / watchlist: totalScore = (genreScore * 0.7 + ratingScore * 0.3) * multiplier
    • Rated-low: totalScore = ratingScore (no genre component)
  5. Deduplicate: Keep the higher score when a movie appears from multiple sources.
  6. Sort by score descending, return top 10.

Response format

Situation Message
No activity "Start exploring movies by rating them or adding them to your watchlist!"
Activity but no qualifying candidates "No recommendations found"
Results "Found N personalized recommendations"

Each recommendation includes: id, title, year, rating, genre, popularity, score, source ("rated" or "watchlist"), sourceMovie, and optionally userRating (only when source is "rated").


What to look for

Each bug is a small, realistic mistake. Think about:

  • HTTP status codes - are the right ones being returned for create vs update?
  • Conditional logic - are the conditions checking the right thing?
  • Query filters - is the quality filter and source exclusion applied everywhere?
  • Data labels - do the response fields match the source of the recommendation?
  • Ordering - are results sorted before being sliced?

Solve it in the browser on GenAIDojo