Transforming Recommendation Systems with LLMs: Architectures, Challenges, and Best Practices

Large Language Models (LLMs) are reshaping the future of recommendation systems—moving beyond traditional models to unlock a deeper understanding of user intent and content semantics. In this session, we’ll trace the evolution from matrix factorization to neural networks, and spotlight how LLMs now power next-item prediction, conversational recommendations, and generative personalization. We’ll then tackle the real-world challenges of deploying these systems at scale—from data bias and model hallucination to inference latency and infrastructure costs—sharing proven strategies to overcome them. Walk away with a clear roadmap for building smarter, LLM-powered recommender systems.

Key Takeaways:

• LLMs are revolutionizing recommendation systems, enhancing understanding of user intent and item semantics beyond traditional models.
• The session traces the evolution from matrix factorization to deep learning to LLM-powered recommendation paradigms.
• Real-world deployment challenges span data biases, model hallucinations, system scalability, and infrastructure costs.
• Actionable strategies and best practices will be shared to tackle issues in data, modeling, and infrastructure for LLM-based recommenders.

Aditya is a seasoned machine learning expert focused on scaling Large Language Models like Llama for recommendation, ranking, and integrity systems. At Meta, he led ML projects for Facebook Reels, driving innovations in user engagement and policy enforcement. With prior experience at Google and as a founding engineer in Area 120, he brings deep practical expertise. A CMU graduate, Aditya is an active voice in the GenAI community, speaking at major conferences and reviewing for top AI journals. You can reach him on LinkedIn.