Technical Writing
Overview
I enjoy documenting engineering ideas as much as building them. My writing focuses on machine learning systems, MLOps, open source, with articles published through community blogs, and technical publications.
This page serves as a curated collection of my published writing, highlighting practical engineering techniques, implementation details, and lessons learned from building production ML systems.
Community Showcase of dagster-hf-datasets
Details the engineering behind the official Hugging Face Datasets integration for Dagster, covering asset abstractions, custom IO managers, metadata, asset lineage, and production-ready dataset orchestration.
Dagster · Hugging Face Datasets · Data Engineering · MLOps
- Community Showcase Part 2
- From Data Repositories to Production Data Pipelines: Bridging Hugging Face Datasets and Dagster with dagster-hf-datasets
Observability with Trackio
The first article explores the design and implementation of an Optuna integration for Trackio. The second expands into a broader observability architecture spanning multiple ML frameworks, discussing design trade-offs, instrumentation, and system architecture.
Trackio · Optuna · Observability · ML Infrastructure
- Designing and Contributing an Optuna Integration That Performs Well in Practice
- Observability for One, Observability for All
Pruna Work
Explores DAG-based model optimization in Pruna, followed by a collaborative article demonstrating how SmolLM can be optimized for faster inference while maintaining model quality.
Pruna · Model Optimization · Inference
- How Pruna Optimizes Models: Tracing the Smash Function and DAG Execution
- SmolLM: Tiny Giants Optimized for Speed
Hyperparameter Optimization with Optuna and Transformers
Demonstrates how Optuna integrates with Hugging Face Transformers to automate hyperparameter optimization while maintaining reproducible machine learning experiments.
Transformers · Optuna · Hyperparameter Optimization · NLP
Model Optimization: Project Deep Dive and General Lessons
One article examines a model-first optimization project and the engineering decisions behind it. The companion article distills broader lessons on building efficient AI inference systems across multiple production deployments.
Model Optimization · Inference Efficiency · Deployment · ML Infrastructure
- Optimizing Where It’s Safest: A Model-First Approach
- Five Deployments In: Lessons on Efficient AI Inference at Scale
Open Source Work
Reflects on recurring software engineering patterns observed across large-scale open-source contributions, alongside a companion article describing the design of an automated contribution tracking and changelog system.
Open Source · Software Engineering · ML Infrastructure
- What 90+ Open-Source PRs Taught Me About Software Quality and Engineering at Scale
- From Commits to Impact: Building an Automated Changelog for Open Source Contributions
HF Buckets Article
Presents a tokenizer benchmarking pipeline built around Hugging Face Storage Buckets, comparing fifteen tokenizers on identical input to highlight practical differences in tokenizer behavior.
Tokenizers · Benchmarking · Hugging Face Storage Buckets
Pixel Art Bench
Introduces a benchmark dataset and evaluation pipeline for pixel art generation, covering dataset construction, exploratory analysis, benchmarking methodology, and fine-tuning experiments.
Benchmarking · Evaluation · Fine-Tuning · Hugging Face