WatchWorthy: A Thoughtful Movie Recommendation Engine

WatchWorthy is a Netflix-style AI movie recommendation app. Its core feature is using multi-step reasoning agents to analyze users' emotions, time budgets, and viewing history, providing personalized recommendations like a movie-savvy friend. It is an entry in the Creative Application Track of the Microsoft Agents League Hackathon, aiming to solve the "black box" problem of traditional recommendation systems and make the recommendation process transparent and explainable.

Traditional streaming recommendation systems are "black boxes"—they only provide "might like" results without explaining the reasons, and ignore users' actual situations (key information like time, companions, emotions, etc.). WatchWorthy was created to address this pain point by treating recommendations as a multi-step reasoning task rather than a simple similarity match.

The core of WatchWorthy is an explicit, ordered reasoning agent that returns a complete reasoning chain for users to view. The steps include: 1. Analyze the user's emotion and available time; 2. Cross-reference viewing history (to avoid repeated or rejected films); 3. Filter the film library by emotion tags and duration; 4. Score candidate films based on preferences; 5. Select primary and alternative recommendations; 6. Write a personalized explanation (citing user information, like a friend's recommendation).

Dual Backend Architecture: Supports Claude (claude-sonnet-4-20250514, main brain) and GitHub Models (openai/gpt-4o-mini, alternative), and users can switch between them.

Reliability Guarantees: Fully localized movie dataset (stable); offline fallback mechanism (uses local reasoning engine when there's no API or call failure); recommendation constraints (only recommends films in the dataset that haven't been watched or rejected); graceful degradation to a dark title plaque when the poster returns a 404 error.

The interface pursues a "digital high-end film magazine" style: First-time users go through an entry taste test (emotion, time, recently liked films); the personalized homepage includes curated recommendations, popular picks, critic recommendations, etc.; hovering over a movie card enlarges it to show ratings/synopses/platforms; the "Help me find a movie" flow (3 questions → reasoning → primary/alternative recommendations); post-view feedback (star ratings, reviews, etc., for future recommendations); a profile page (statistics, favorite genres, watchlist, etc.); fully responsive design (mobile drawer, desktop dialog box).

WatchWorthy demonstrates the potential of AI agents in consumer applications: it not only uses AI for recommendations but also makes the process transparent and explainable, enhancing user trust and satisfaction. This "explainable AI" model can be extended to e-commerce (explaining product recommendations), content curation (news/music), education (learning resource recommendations), and other fields, and may be an important design paradigm for the next generation of intelligent applications.