Technical Essays

I write about the things that spark my curiosity. Some posts will be polished, while others serve as a journal for me.
Hopefully you’ll find something here to inspire you too.

Memory Leak in Jupyter Notebooks and How to Fix It

Python
Jupyter
IPython
Debugging

The display hook causing memory leaks? Well that’s a new one.

Mar 7, 2026

Implementing Gradient Boosted Trees from Scratch - LightGBM, XGBoost, CatBoost

ML
Algorithms
From Scratch
GBT
XGBoost
LightGBM
CatBoost

Building a minimal GBT library from scratch to understand what’s actually happening under the hood, and to have the flexibility to experiment with new ideas.

Dec 14, 2025

Github Workflow Tips and Tricks

GitHub
DevOps
CI/CD
Automation

A collection of useful tips and tricks for better leveraging Github workflows

Aug 7, 2025

A Review of ML Time Series Forecasting Models

ML
Time Series
Forecasting
LSTM
N-BEATS
TFT
DeepAR

A review of some of the well-known machine learning time series forecasting models.

Feb 28, 2025

Saving Time: Post-mortem Debugging in Python

Python
Debugging

A bit of Python black magic that lets you efficiently inspect and manipulate execution contexts after crashes, aka, post-mortem debugging.

Jan 30, 2025

Running Llama 3 in the Browser!

NLP
LLM
WebLLM
MLC-LLM
WebGPU
TVM

Chat with Llama and other LLMs that run locally in your browser using WebLLM.

Oct 6, 2024

Information Theory Fundamentals: Entropy, Cross-Entropy, and KL Divergence

ML
Information Theory

A guide to information theory covering entropy, cross-entropy, KL divergence, and Jensen-Shannon divergence with proofs, examples, and visualizations.

Aug 24, 2024

Custom Loss Functions for LightGBM and CatBoost

ML
LightGBM
CatBoost
GBT

Guide on how to implement custom loss functions and evaluation metrics for LightGBM and CatBoost

Aug 11, 2024

Implementing Multi-Layer LSTM and AdamW from Scratch using NumPy

ML
NLP
Optimization
Analysis
From Scratch

This tutorial walks through the implementation of a multi-layer LSTM model from scratch in pure NumPy, and trains it on the Shakespeare dataset. It also covers the implementation of the AdamW optimizer and the necessary data modules.

Jun 15, 2024
No matching items