I’m an independent software architect at Redwood Consulting and a Google Developer Expert for Cloud. This site indexes everything I write—original essays published here, Binary Alchemy on Substack, and engineering notes across the web.
-
Google Jules Hands Off to Antigravity
The background coding agent retires, and its best idea moves on
Jules is retiring. Google announced the turndown on October 29, Jules went read-only on November 11, and on December 31 it's gone. What Google learned from…
-
Sidekiq to Temporal: a zero-downtime migration strategy
Outlines a 'Wrap, Refactor, Reroute, Retire' strategy for migrating high-availability production systems from Sidekiq to Temporal. It emphasizes using a compatibility context to ensure logic consistency…
-
LLMs, Requirements, and Results
LLMs simplify the translation of ideas into code, but the fundamental challenge of accurately specifying requirements remains unchanged. Original post on Substack
-
Deploying to Firebase Hosting + Firestore from GitHub actions
Provides a guide and YAML configuration for automating Firebase deployments using GitHub Actions. It covers setting up environment variables, service accounts, and permissions for Hosting and…
-
Quiz: Ruby & Rspec scoping
Presents a short quiz about Ruby and RSpec variable scoping within test blocks. It challenges readers to predict whether a specific test case involving let and…
-
Performance trap: general libraries & helper objects
Explores how general-purpose library abstractions can introduce significant performance overhead due to unnecessary object construction and attribute handling. It uses a cellular segmentation example to show…
-
Tidy First?, reflect next: chapter 6
"Cohesion order" is the practice of colocating code that changes together to surface coupling and simplify future behavior changes. Original post on Substack
-
Tidy First?, reflect next: chapter 5
Reordering file contents improves readability for new readers, despite a personal preference for navigating code via IDE tools. Original post on Substack
-
Optimizing QuPath intensity measurements: 12.5 hr to 2min
Describes a massive performance optimization in spatial biology image analysis by using the Proxy pattern to prefetch images into memory. This change reduced the time for…
-
Tidy First?, reflect next: chapter 4
Wrapping old implementations with new interfaces facilitates refactoring, such as converting stateful classes into stateless functions. Original post on Substack
-
Re-rebuilding TF2.8 image: 369 patches
Reports on a maintenance update for a TensorFlow 2.8 Docker image that incorporated 369 security fixes from upstream Ubuntu changes. It also mentions the challenges of…
-
Tidy First?, reflect next: chapter 3
Normalizing symmetries in code is important to manage the "organic" mutation of software and prevent creativity from devolving into chaos. Original post on Substack
-
Tidy First?, reflect next: chapter 2
The proactive removal of dead code—code that never runs or does nothing—is an explicit and valuable step in the software development process. Original post on Substack
-
Tidy First?, reflect next: chapter 1
Applying the concept of guard clauses to simplify complex conditional logic, reflecting on Kent Beck's "Tidy First?". Original post on Substack
-
gs-fastcopy: get CPU count for upload workers
Discusses an enhancement to the gs-fastcopy package that automatically detects the available CPU count to set optimal worker defaults. This improvement helps maximize performance on multi-core…
-
Introducing gs-fastcopy
Introduces a Python package designed to optimize Google Cloud Storage file transfers by using multi-threading. It aims to provide a faster alternative to the standard single-threaded…
-
Ensuring GCE instances have full access to GCP APIs
Explains that default Google Compute Engine settings restrict API access even if the service account has permissions. It provides a guide on how to enable 'full…
-
Improve TensorFlow model load time by ~70% using HDF5 instead of SavedModel
Demonstrates how switching from TensorFlow's SavedModel format to HDF5 can significantly reduce model load times. This optimization is particularly beneficial for serverless environments where models must…
-
GitHub to Artifact Registry & Docker Hub via Cloud Build
Details the process of automating container builds using GitHub Actions to trigger Google Cloud Build. The resulting images are pushed to both Google Artifact Registry and…
-
One Byte Explainer: Large Language Models
Explains that LLMs communicate by modeling human patterns from vast training data. It notes their limitations, such as relying solely on training patterns and their tendency…
-
One Byte Explainer: Recursion
Describes recursion as a problem-solving technique that splits complex problems into simpler, identical pieces. It emphasizes the 'divide and conquer' approach to solving anything by reducing…
-
One Byte Explainer: Big O notation
Explains Big O notation as a way to represent how code runtime grows at scale. It uses the example of handshakes among people to illustrate exponential…
-
One Byte Explainer: the Halting Problem
Explains the Halting Problem using Alan Turing's 1936 proof that we cannot know if a loop will eventually finish. It highlights the paradox of a program…
-
Running DeepCell on Google Batch with node pools
We benchmarked running DeepCell on Google Batch and found that using node pools reduced runtimes by 80% for large containers. Original post on Medium
-
Using TensorFlow 2.8 on an Apple Silicon arm64 chip
Due to dihydrogen monoxide interfacing with my computer, I bought a new Mac. That meant going from x86 to arm64, causing some shenanigans. Original post on…
-
Rebuilding TensorFlow 2.8.4 on Ubuntu 22.04 to patch vulnerabilities
TLDR: The official TF 2.8.4 container is missing a bunch of patches. Let's rebuild TF 2.8.4 using more recent bases! DeepCell + tensorflow-2.8.4 DeepCell + tensorflow-2.8.4-redux…
-
Google Cloud Build + personal GitHub repos?
Question: how to connect Google Cloud Build to personal GitHub repos with a service account? (spoiler alert: you can't) Original post on dev.to
-
Container size analysis: TensorFlow 2.8 base image vs Deep Learning
TLDR, building our DeepCell container from a base TensorFlow image is 50% faster to load and 60% smaller than using the Deep Learning container. Deep Learning…
-
First impressions: GPU + GCP Batch
We tried (and failed) to run GPU jobs using Google Batch. Original post on dev.to
-
Algorithm edge cases vs notation
I spent 2.5 hours debugging an infinite loop in this algorithm, caused by an edge case (center point of structuring element is excluded). I'd missed a…
-
reading: Wrangling a Python monorepo
Been a long time since I've wrangled a Python deployment let alone a Python monorepo. I found this series by Tweag.io invaluable: PYTHON MONOREPO: AN EXAMPLE…
-
Running a Vertex AI custom container
First steps running a Vertex AI custom container job. Original post on dev.to
-
Benchmarking project update: new issue views
Providing better visibility into issue status. Original post on dev.to
-
How to debug Cython C?? 🤔
I'm writing Cython, but how do I debug it? Original post on dev.to
-
Building a Vertex AI custom job container
First steps creating a Vertex AI custom container. Original post on dev.to
-
Using chaos to build confidence in software rewrites
A little chaos can create a lot of certainty. Original post on dev.to
-
The cost of types
Even though types are great, there's no free lunch. Original post on dev.to
-
Language types for integration safety
How language types provide built-in integration safety. Original post on dev.to
-
Kotlin in the browser
Explorations using Kotlin to write web apps. Original post on dev.to
-
Get GPU names + counts with TensorFlow API
How to get information on available GPUs, using the TensorFlow API. Original post on dev.to
-
Getting notebook's machine type on Vertex AI
How to get the active machine type, from a Vertex AI notebook. Original post on dev.to
-
Home wireless fundamentals
Explains the theoretical and practical aspects of Wi-Fi, including how radio frequencies and international standards like 802.11 determine maximum speeds. It highlights the difference between theoretical…
-
Home internet fundamentals
Introduces the basic components of a home network, including routers, modems, and ISP connections. It uses a personal troubleshooting story to explain how to drill down…
-
DeepCell Notes 20230701
Notes when getting started
Saving these for later as I get into DeepCell– DeepCell Docs DeepCell online prediction I found this tool useful/essential for viewing TIFF files from wet lab…
-
Hello, MNIST
Serverless machine learning training & prediction
How does AI work in the cloud? My friend Lynn encouraged me to learn about bioinformatics , the convergence of Software (informatics) and Life Sciences (bio).…
-
Hello, Rosalind
Learning about bioinformatics
I like learning new things. My friend Lynn encouraged me to learn about bioinformatics , the convergence of Software (informatics) and Life Sciences (bio). I know…
-
The 80% abstraction
Proposes a method for simplifying large software monoliths by focusing on abstractions that cover 80% of use cases rather than aiming for 100% perfection. It uses…
-
Who knows about the overloaded data?
Reducing complexity to scale software
In evolutionary architecture, data models get overloaded – inevitably. Some new requirement comes in and we introduce some property that changes how the data is interpreted.…
-
Refactoring: Inside-out naming
Name in the context of now
Naming should be consistent with today’s context – not yesterday's or yesteryear’s. The function extraction refactor is a great opportunity to refine naming. Usually we’re extracting…
-
Migrating passwords from Google to 1Password
Enhancing personal security stance
I've moved my "saved passwords" from Google passwords to 1Password . In other words, I've moved my secret password vault from Google's business into 1Password. For…
-
2020 election: USA/CA/Oakland
A franco-american guide to politics
Official CA Voter Information Guide Federal Government President and vice president Biden/Harris. It almost doesn't matter who's running against Trump. We'll fix it later. Those who…
-
Booting Home Assistant
Out-of-box to online RasPi4
I set out to build my own smart home controller after my previous hub's manufacturer switched to subscription on one week's notice. I chose to run…
-
Selecting a smart home platform
Surveying open-source home automation
My home came with Z-Wave switches controlled by a Wink2 Hub. Z-Wave is radio for home automation devices. Wink is a smart home platform; the hub…
-
Everyday Math: bulk discounts
Uses algebra to analyze whether annual subscription discounts are actually worth the upfront cost based on expected usage duration. It provides a mathematical formula to determine…
-
Serverless Conversation with Google DialogFlow
A tutorial on building a conversational agent using Google DialogFlow and Cloud Functions to track household routines. It covers creating intents, training phrases, and extracting entities…
-
Economics of freelancing
Delivering value as an independent contractor
As a freelancer, you’re hired to save your client money or time (but 💰 = ⏳). So let’s just call it work. How much work you…
-
JAMstack: Angular7 + MaterialDesign + Bootstrap + Netlify
A serverless material-design application
A JAMstack is a web front-end delivered from a CDN, not from a back-end service (e.g., Ruby on Rails, Node, Golang, etc.). Dynamic content is rendered…
-
What's in a name? Tools for allyship
Updating from FactoryGirl to FactoryBot
On October 24th, 2017 the ThoughtBot team renamed their popular Ruby testing library FactoryGirl to FactoryBot . The change, and even the request itself, was controversial…