2025 Year in Review

2025-12-14 · Source: Eugene Yan · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Software Development & Engineering · Depth: Novice, short

Summary

Ziyou Yan's "2025 Year in Review" details significant personal and professional achievements, including doubling down on health, career advancement, and two vacations. Professionally, Yan successfully published six substantive pieces, bridging RecSys and LLMs, and prototyped four applications, notably the AI Reading Club (ARC) and an LLM-RecSys Hybrid. The LLM-RecSys Hybrid demonstrated a novel capability to recommend items based on natural language or item ID input, even across platforms like Nintendo and Xbox. Health goals, such as weekly 5km runs, showed gradual improvements in resting heart rate and VO2 max. Career milestones included a promotion to Principal Applied Scientist and O1 visa approval. Site statistics for 2025 indicated a 71% year-over-year increase in Google search impressions to 3.39M, but a 25% decrease in clicks to 42.2k, with a corresponding drop in CTR from 2.9% to 1.2%, possibly due to AI Overviews.

Key takeaway

For AI Engineers and Research Scientists exploring next-generation recommendation systems, you should investigate the LLM-RecSys Hybrid approach. This method allows for steerable recommendations using natural language and semantic IDs, potentially offering a significant advancement over traditional behavioral models. Consider prototyping similar hybrid applications to leverage LLM world knowledge for more intuitive and flexible item suggestions, especially for cross-platform or genre-specific filtering.

Key insights

Integrating LLMs with RecSys can yield powerful, steerable recommendation systems capable of cross-platform suggestions.

Principles

• Consistent effort improves health metrics.
• Bridging distinct technical fields creates novel solutions.

Method

Combine LLM world knowledge with behavioral RecSys for steerable recommendations, training models to natively recommend items via natural language or item ID input.

In practice

• Explore LLM-RecSys hybrids for enhanced recommendations.
• Utilize semantic IDs for steerable recommendation systems.
• Monitor search metrics for AI Overview impact.

Topics

• LLM-RecSys Hybrid
• Recommendation Systems
• AI Engineer World's Fair
• AI Overviews
• Principal Applied Scientist

Best for: AI Scientist, AI Engineer, Machine Learning Engineer, Research Scientist

Related on AIssential

• Semantic Search At LinkedIn, LLM-Driven Autonomous Optimization for Industrial-Scale Recommendation Systems, and More! — Recent advancements in information retrieval focus on LLM-driven optimization, efficient embeddings, and scalable recommendation systems.
• The Future is Agentic in Recommender Systems — LLMs are transforming recommender systems from ranked lists to agentic, task-oriented solutions, while introducing new trustworthiness challenges.
• Increase Recommendation Systems’ Precision with LLMs, Using Python — Optimize LLM-powered recommendation systems by combining cheap, high-recall filtering with expensive, high-precision reranking.
• How Amazon Uses LLMs to Recommend Products — Amazon's COSMO system uses LLMs and a multi-stage pipeline to build and serve commonsense knowledge for e-commerce.
• Personalization in the Era of LLMs - Shivam Verma, Spotify — Spotify is integrating LLMs into its recommendation system using user embeddings, semantic IDs, and soft tokens for enhanced personalization.
• IEEE Transactions on Artificial Intelligence, Volume 7, Issue 2, February 2026 — The volume showcases diverse AI advancements, from LLM applications in autonomous driving to specialized models for protein language processing.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Eugene Yan.