Hello world
Why this blog exists, the evidence-first mindset, disclaimers, and what to expect.
This blog is my public lab notebook. I pursue evidence: clear questions, controlled setups, benchmarks when useful, and honest takes on what works where.
No post pretends to be complete, final, or globally true. I learn as I go, jot what I figure out, stay ready to correct mistakes, and admit blind spots when readers see angles I missed.
Unless I flag otherwise, posts mirror my wording on my gear. For biography, PGP keys, and mail etiquette, read About.
Disclaimers
Personal opinion only: not my employers', clients', or sponsors'. When a topic touches work I still split personal takes from anything official.
I fund the experiments: my machines, cloud credits I pay for, and code I control unless a post states sponsored GPU time or borrowed kit. I surface that at the top when it happens.
Spotted a mistake or disagree? Use [email protected]. That is the inbox I read for substantive replies.
The mantra this site keeps coming back to
I care about answers you can stress-test.
When that is realistic I write close to the scientific method: state the claim, fix the scenario, compare options with benchmarks or a small reproducible demo, then say what wins in this setup (not universal slogans). Where evidence stays thin I say so and file the piece as a notebook entry, not a final verdict.
A lot of why I publish is simple: prose drags assumptions, harness design choices, and failure modes into the open.
Replication fits the same habit. Document procedures clearly enough that readers can rerun them. When calibration matters, rerun external papers or older versions of my own notes. Same curve or a different one, both outcomes teach you something.
Why bother publishing
Writing keeps me honest. Turning fuzzy intuition into text pushes me toward clearer assumptions, tighter experiments, and explicit limits.
This site is also external memory. Across a typical year most fragments never leave local repos or notebooks. A public subset helps me see what actually shipped instead of feeling stuck.
When friends ask how I think through a topic, I send one link that already records trade-offs, failures, or partial wins.
Posting drafts stress-tests ideas. Some posts lock a hunch behind numbers; others age badly and need edits. Both are useful.
When a post is mostly opinion, keeping a dated public copy lets me stay honest about drift: point to older claims, admit when something aged poorly, show where I changed my mind, and explain why without pretending consistency I never had.
What you'll find
- ML engineering: tooling and delivery notes.
- Agentic systems: experiments backed by setups I can describe.
- Reverse engineering: curiosity, written without being reckless.
- Art-adjacent: tangents when they collide with the same stack.
- Projects: builds I maintain or poke at seriously.
- And maybe more someday. đ
Want plain Markdown instead of HTML? Swap .html for
.md on the URL path.
For this note that's /2026/05/01/hello-world.md
(front matter included). Handy if you want to edit offline or paste into an agent.