Artem Demchyshyn

Java backend engineer building document-generation systems, developer tools, and Al-backed workflow products.

[Artem Demchyshyn @DemchaAV]

GraphCompose v1.8 "illustrative" 🪄

A declarative Java engine for business PDFs — and the banner above? Rendered by the engine itself.

→ native vector charts (bar / line / pie)
→ SVG import + gradients
→ leaner jar: ~20 MB → ~2 MB
Java DSL. Author intent, not coordinates.
#Java #OpenSource #PDF #DataViz
⭐ https://t.co/Am22KSlCMv

[Artem Demchyshyn @DemchaAV]

I've just updated my Google Pixel 10 Pro XL with new Android 17. What did I have updated? I have added bubbles to my phone. Thanks!

[Artem Demchyshyn @DemchaAV]

@Gencoin8 To be fair, the branch marks always said that model is the best. But when you start using it, you realize it's not. It does some things better, but always loses control and doesn't follow your instructions.

[Artem Demchyshyn @DemchaAV]

@Web3_Jarin Hello

[Artem Demchyshyn @DemchaAV]

https://t.co/DYjHw4wd78

[Artem Demchyshyn @DemchaAV]

@dotgil You are welcome 🫠

[Artem Demchyshyn @DemchaAV]

GraphCompose v1.7.0 — "geometric" — just shipped to Maven Central.

implementation("io.github.demchaav:graph-compose:1.7.0")

Geometry is now a first-class authoring primitive 🧵

[Artem Demchyshyn @DemchaAV]

@xikhar It's fork from Fable 5 😆

[Artem Demchyshyn @DemchaAV]

@pankajkumar_dev @float_notes I like that 🤣

[Artem Demchyshyn @DemchaAV]

@Amank1412 Unfortunately it seems like Sam Altman dreams 😅

[Artem Demchyshyn @DemchaAV]

@RoundtableSpace Wow very useful, especially when you have no access to it 🤣

[Artem Demchyshyn @DemchaAV]

Yesterday, Anthropic released Claude Fable 5, a Mythos-class model that looks like an attempt to give broader public access to something above the Opus level — but with a much stricter safety layer around it.

I’ve only tested it briefly, so this is not a deep review yet. But my first impression is strong. The model feels thoughtful, capable, and very good at holding context. In some complex tasks, it even felt stronger than Opus.

What interests me most, though, is not just the raw intelligence.

Some responses feel like they go through an extra safety process: deeper thinking, a pause, then a final answer that sometimes appears in chunks rather than as a smooth token stream. My guess is that Fable is not just using a simple input/output filter. It feels more like a multi-layer system: detection, safety gates, possible rewriting, and maybe fallback to a more restricted model when the request becomes sensitive.

That makes Fable interesting not only as a new model, but as a public experiment in how frontier AI might be deployed:

powerful model inside, safety layers outside, monitoring, fallback, retention, and pricing as an additional filter.

The pricing is also worth watching. Anthropic says the model should be more token-efficient, but it costs more than Opus. So the real question is not simply “is it smarter?”, but “does it solve hard tasks in fewer iterations?”

For everyday prompts, probably not worth it.

For complex reasoning, coding, or long-context work — maybe.

Either way, Fable feels like a glimpse of where frontier AI access is going: more powerful, more controlled, more expensive, and slightly uncomfortable.

Exciting and a little scary at the same time.

#AI #ClaudeAI #Anthropic #AISafety #LLM

[Artem Demchyshyn @DemchaAV]

On Maven Central (Java 17+, MIT):

implementation("io.github.demchaav:graph-compose:1.7.0")

Release: https://t.co/RNtIbHcOHx
Showcase: https://t.co/H8FcknNhP5

Author intent, not coordinates.

Bookmarks help.

[Artem Demchyshyn @DemchaAV]

Zero breaking changes from 1.6.x — bump and rebuild.

1144 tests green. Deterministic two-pass layout = snapshot-stable renders. GPG-signed on Central, with a binary-compat gate on every PR.

[Artem Demchyshyn @DemchaAV]

@ClaudeDevs I was thinking, why is it going so fast

[Artem Demchyshyn @DemchaAV]

Install:

implementation('io.github.demchaav:graph-compose:1.6.7')

Or Maven: io.github.demchaav:graph-compose:1.6.7

Release notes: https://t.co/BKuCfegK8b

Bookmarks help.

[Artem Demchyshyn @DemchaAV]

GraphCompose v1.6.7 just hit Maven Central.

A patch release. Zero breaking changes. ~30 fewer jars on your classpath if you used the prior version.

🧵 What got cleaned up.

[Artem Demchyshyn @DemchaAV]

japicmp gate vs the v1.6.6 baseline: "semver PATCH, compatible bug fix."

The one surface delta is NodeRegistry going non-final so the session can install an auto-invalidating subclass.

Every existing public call site compiles + runs unchanged. Migration: nothing for typical usage.