I Built a Movie Recommendation Agent to Solve Movie Nights with My Wife (rokn.io)

A developer recently unveiled a personalized movie recommendation agent, dubbed movieagent.io, designed to ease the common struggle of selecting films for couples. Faced with diverging tastes—one partner favoring romantic comedies and the other critically acclaimed sagas—this multi-agent system leverages LLMs and vector mathematics to facilitate an interactive movie selection process. Users engage in a dynamic conversation with the agent, which utilizes a series of categorical and duel questions to gather preferences. The system combines a main orchestration agent, powered by Claude Sonnet 4.5, with a search agent using Haiku 4.5, enhancing efficiency and recommendation accuracy by separating conversational management from search tasks. The significance of this approach lies in its innovative architecture and user-friendly design, which avoids complex text inputs that can stifle conversation flow. The search agent employs embeddings to better represent movie qualities beyond basic descriptions, enhancing the diversity of recommendations—a key challenge in traditional LLMs. By including augmented data descriptions, the agent captures the essence of films, which fuels the recommendation process. This project not only provides a practical solution for couples but also contributes to the AI/ML community by demonstrating effective strategies for developing more nuanced, user-aligned recommendation systems.