Skip to main content
JHJames Hu
HomeBlogPortfolioBookAboutContact
HomeBlogPortfolioBookAboutContact
James Hu·ML systems & production architecture·© 2026
Portfolio·Blog·Book·About·Contact·Privacy

Blog

Writing about system design, ML systems, and building production-grade software.

Featured Series : Architecture, Systems, and Intelligence

Essays on how ML systems behave under real-world constraints.

commentary

The C++ Renaissance

How AI Collapses and Rewrites the Constraints That Created Modern Languages.

c++
programming-languages
ai
+6
architecture

I Don't Debug Code — I Debug Agreements

Systems fail at boundaries, not implementations. A practice for writing falsifiable contracts and debugging by auditing agreements instead of tracing execution.

debugging
architecture
contracts
+5
All Posts
Explore topics:
ai (3)
ai-engineering (3)
book (3)
llm (3)
genomics (2)
history-of-science (2)
language-models (2)
machine-learning (2)
Cover of the book AI Engineering: Building Production-Ready LLM Applications by James HuConcept

How to Think About AI Agents

A working mental model for agents — and when not to build one. A companion to Chapter 8 of my book, AI Engineering.

agents
agentic-systems
mcp
tool-use
+3
May 29, 20262 min read33
Cover of the book AI Engineering: Building Production-Ready LLM Applications by James HuConcept

RAG Is an Architecture, Not a Feature

Why retrieval-augmented generation is a system you design, not a library you import — a companion to Chapter 7 of my book, AI Engineering.

rag
embeddings
vector-search
reranking
+3
May 29, 20262 min read31
Cover of the book AI Engineering: Building Production-Ready LLM Applications by James HuConcept

Building Your First LLM Application

The gap between calling an API and shipping an app — a companion to Chapter 4 of my book, AI Engineering.

llm
ai-engineering
rag
getting-started
+2
May 29, 20263 min read25
A DNA double helix rendered as a glowing blue wireframe network of nodes and edges on a dark background, evoking both a genome and a neural networkExploration

Neural Networks Got Too Big to Read - Now We Sequence Them

Why large language models stopped reminding me of software and started reminding me of genomics — and what that says about studying any system too large to read.

interpretability
machine-learning
genomics
neural-networks
+6
May 28, 202610 min read42
Abstract sphere of interconnected nodes evoking attention edges between tokens in a neural networkExploration

The other ninety-eight percent

What the exome era of genomics reveals about subquadratic attention — and the belief trap that follows a good, necessary compromise.

attention
transformers
ai
language-models
+5
May 26, 20265 min read67
Abstract circuit board patterns representing the intersection of programming languages and AICommentary

The C++ Renaissance

How AI Collapses and Rewrites the Constraints That Created Modern Languages.

c++
programming-languages
ai
language-design
+5
Jan 19, 20267 min read44
Vintage navigation map representing the agreements that guide system architectureArchitecture

I Don't Debug Code — I Debug Agreements

Systems fail at boundaries, not implementations. A practice for writing falsifiable contracts and debugging by auditing agreements instead of tracing execution.

debugging
architecture
contracts
agreements
+4
Dec 15, 202511 min read60
DNA double helix structure representing the intersection of biology and computingExploration

Why Reading Backwards Is Hard: Endianness in Computing and Directionality in DNA

An exploration of the surprising parallels between byte order in computer systems and strand directionality in molecular biology, revealing how both domains solve the same fundamental problem.

endianness
dna
bioinformatics
computer-science
+2
Dec 10, 20255 min read30