CUDA Performance, Part 1: Device Busy Is Not GPU Busy

A program can report 98.8% device utilization while leaving its compute units idle 39% of the time. This is what I learned building a TUI for nsys timelines, and why the second number is the only one that matters.

Bitcoin Regime Analysis, Part 2: What Each Feature Buys You

Volatility buys long regimes with no direction. Volume buys direction but halves regime length. Fractional differencing buys length back and destroys direction completely. Every one of those is a trade, not an improvement.

Bitcoin Regime Analysis, Part 1: What the Data Looks Like Before You Model It

Bitcoin’s returns are almost unpredictable in the mean, wildly structured in the variance, and nothing like a normal distribution. This is a walk through the checks that establish that, and why they point at Hidden Markov Models.

Hello, welcome, and stay tuned!

Welcome to my personal blog! Here I’ll share insights about NLP and Machine Learning, from beginner to advanced topics, along with career growth advice and personal development tips for building a successful career in tech.

Indonesian Language Model

A deep dive into building state-of-the-art Indonesian language models using AWD-LSTM architecture. This post covers the challenges of working with low-resource languages, implementation details, and experimental results from production systems at Traveloka.