$ whoami
Software Engineer · platform engineering × AI-native architecture · real-time UIs · calisthenics as the lens
This is a knowledge garden, not a blog. Notes start as seeds 🌱, grow 🌿 and, with enough tending, become evergreen 🌳 — all linked, the way I actually think.
When anyone with an AI model can decompile a game, root a TV or unlock a GPU feature, code stops being a moat. Expect more lock-down (DRM, streaming, server-side logic), and value moving to what can't be copied: operations, trust, data, licensed IP. Prompted by reports of Claude cutting off reverse-engineering tasks, and DLSS frame generation unlocked on RTX 30 cards.
A one-page context brief I paste into any AI coaching chat: profile, skill baselines vs 12-month targets, tendon guardrails, a mornings-first weekly split and 6-week waves. Update it after each max test.
Once the beginner routine stops working, alternate blocks that build capacity with blocks that turn it into skill, with a deload between. You can't push both at once.
Four years and 168 commits of dotfiles: GNU Stow mirroring $HOME, a Linux-to-macOS migration, a graveyard of Neovim configs, scripts that encode my workflow, and keyboard firmware as config too.
In 2019 I fixed failing tests from a train with Termux. In 2026 the phone doesn't need to do the work at all: I SSH home and steer a coding agent from my pocket, with a cloud box and notes-as-triggers next.
Flyweight = share what's common, pass in what varies. Applied to byte buffers: one reusable view object reads fields straight from the bytes (no per-message objects). For JSON with a repeated shape, build the shape once and share it; each message gets only a small index of offsets. It's Rust's borrowing, done by discipline instead of the compiler.
Golden paths, internal developer platforms, delivery metrics, and what changes when AI agents become users of the platform too.
62 notes🌿 26🌱 36
Machine learning fundamentals, AI agents, and the research I did at university: topic modelling, social network analysis and online communities.
32 notes🌿 11🌱 21
API design, services and data, messaging and migrations, Node.js and the JVM: the server side of the systems I build, and where I'm growing next.
30 notes🌳 3🌿 12🌱 15
Interfaces that stay smooth under a firehose. Streaming transports, 60fps rendering, state that doesn't thrash, and connections that heal themselves.
20 notes🌳 2🌿 15🌱 3
Component architecture, React patterns, state, performance and the build tooling that keeps a front end fast and easy to change.
42 notes🌳 4🌿 26🌱 12
Testing strategies teams actually adopt, mocking at the network, end-to-end confidence, shifting left, and the security habits that belong in the same pipeline.
30 notes🌳 4🌿 12🌱 14
Failure, consistency, consensus and delivery guarantees. The fundamentals underneath every platform and real-time system, written down so I actually understand them.
36 notes🌳 4🌿 19🌱 13
The environment I build to work fast: dotfiles, provisioning with Ansible, git worktrees, a keyboard-first setup, and tools that let me ramp up anywhere.
25 notes🌳 1🌿 15🌱 9
Second brains, spaced repetition and the daily loop that turns scattered notes into knowledge I can actually use. This site is its visible end.
21 notes🌳 2🌿 12🌱 7
Leading UI teams, running communities of practice, communicating with stakeholders, and the habits that make a team better at its craft over time.
57 notes🌳 2🌿 26🌱 29
A calisthenics skills library, training principles, and the engineering lessons hiding inside them.
34 notes🌿 21🌱 13
Case study — knowledge transfer as a continuous habit: structured sessions backed by written docs, a reuse strategy and a clean backlog, so a team can own a platform without depending on any one person.
Case study — writing the business case to host four career-change students on a real cross-department component library, then mentoring them as line manager and product manager. Three moved into engineering roles. What worked: real work, stretch tasks in the area each person wants, pairing that hands back control, and expectations written as a progression plan.
Case study — building a broadcaster's streaming app for many TV platforms at Accedo. Memory and performance work for slow devices, a clear, agreed back-navigation model, a full rebrand, and a knowledge-sharing culture.
$ nowOct 2026
Life. I became a dad this year, and it has changed how I spend my time. With an office five days a week, close to two hours' commute each way and longer hours, time comes in smaller blocks. So I get up early, keep the priority goals few and get things done effectively. I still train six days a week, and weekends are for my family.
Work. Owning front-end development as a software engineer at a fintech building real-time crypto trading interfaces. Most of my thinking goes into keeping fast-moving UIs smooth and predictable: decoupling rendering from the data feed, holding 60fps under a firehose of updates, and making click-to-trade flows feel instant without cutting corners on quality.
Thinking about. Putting coding agents on a leash. As agents write more of the code, the hard part moves from writing to verifying: harnesses of skills, tests and linters that let an agent correct itself before a person reviews, safe paths into large codebases, and brownfield work where the code is older than the tools. Underneath it all is the platform question I keep coming back to: what a golden path looks like when people and agents both walk it.
$ radar
29 items · 12 adopt · 4 trial · 8 assess · 5 hold
My tech radar →$ ./lab --train
A library of 42 calisthenics skill families: 131 progressions unlocked, 18 current goals — and the engineering lessons hiding in progressive overload, deloads and balance.
Open the skills library →