Deep Dive Series

Recommendation Systems

How recommendation systems work at scale — from candidate retrieval to ranking, multi-objective optimization, and generative RecSys.

[RecSys] Part 1: Intro and common blind spots

[RecSys] Part 2: Two tower models in industry

[RecSys] Part 3: LONGER, scaling up long sequence modelling in industrial recommenders

[RecSys] Part 4: Real time bandits at YouTube

[Bonus] Pinterest Recommendation Systems evolution through the years: deep dive of 7 architectures

[RecSys] Grand final!

LLM Inference

How large language models are served in production — from attention mechanics to batching, quantization, and serving infrastructure.

LLM serving (1): Continuous batching

LLM Serving (2): Paged attention

LLM serving (3): Speculative decoding

LLM Serving (4): Disaggregated serving

LLM Serving (Bonus!): takeaways from industry