Blog
Writing on machine learning, recommender systems, and the occasional side quest.
- Misc
My Son's First Gaming Console is Retro (No Internet, Ads, or Predatory Mechanics)
In a study of 135 apps marketed to under-fives, 95% carried ads. So my son's first console is a $45 setup: a retro handheld with no wifi, no ads, and only the games I pick.
- Misc
Completing the Pokédex (in TCG)
With the help of Claude Code, I completed a correctly-ordered list of all 1,025 Pokémon (National Dex order).
- Projects
Building an LLM-Powered Search and Discovery Experience for a Movie Recommender
LLMs are becoming a core part of how industry recommendation systems work: Netflix's natural-language search, Spotify's AI playlists. This project brings that same idea to movies.
- Projects
Recency Is All You Need: A 12-Model Experiment on User-History Representation for Two-Tower Retrieval in Recommenders
The textbook move in deep two-tower retrieval models is to pool a user's interaction history into a single input vector. It's the standard approach, and it buries possibly the most predictive signal there is: the last item the user interacted with.
- Projects
Taming popularity bias in recommender systems with logit adjustment
My two-tower recommender kept handing everyone the same blockbusters. Here's how one term in the loss got it surfacing the niche films that actually fit each user's taste.
- Projects
$200 vs $200k: Generating Item Features With an LLM Instead of Hand Tagging
A controlled two-tower recommender experiment on MovieLens 32M and what it means for a team setting up recommendations with no tagging budget.
- Projects
Building a Two-Tower Deep Learning Movie Recommender System in Pytorch (from scratch)
Building a two-tower deep learning movie recommender system from scratch in PyTorch, end to end.
- Misc
35 Books that Changed My Life
35 books that changed my life, with a few words on how each one did.
- Projects
Implementing Algorand Agreement
Implementing the Algorand Agreement protocol — fast, fork-free Byzantine consensus for a scalable cryptocurrency.
- Projects
GPU Accelerated Matrix Factorization for Recommender Systems
cu2rec: a CUDA matrix-factorization library for recommender systems that runs parallel SGD up to 10x faster than state-of-the-art sequential implementations.
- Misc
Concurrency With Go
A quick guide to writing concurrent function calls in Go with goroutines and channels.
- Projects
Building Books2Rec - A Book Recommender System
Building Books2Rec, a book recommender system using SVD and autoencoders on Goodreads and Amazon data.
- Misc
The Effective Engineer - TLDR
A condensed TL;DR of Edmond Lau's 'The Effective Engineer' — the highest-leverage ideas, in my own words.
- Misc
Recommender Systems Overview
Slides from a presentation I gave on recommender systems for my NYU Big Data Science class.
- Misc
Learning How to Learn
An overview of Coursera's 'Learning How to Learn' — chunking, memory, focused vs. diffuse thinking, and beating procrastination.
- Misc
Recommender System with Rapid Miner
A step-by-step guide to prototyping a user-kNN collaborative filtering book recommender in RapidMiner.
- Misc
50+ Data Science Terms
A glossary of 50+ data science and machine learning terms, defined for an NYU Big Data Science class.