§ BLOG

Blog.

Notes on development, infrastructure and AI. What was built, where, and how.

13 May 2026

Fail2ban archaeology: five traps that quietly piled up dead bans for years

Went into Grafana to see who's hitting the site — found 1,300 self-hits per day. Pulled the thread: a broken action, a regex banning the host header, 1,339 "banned" IPs with no iptables rule. Trust but verify.

#fail2ban#security#nginx#monitoring#devops

12 May 2026

20 CPU/tick: how I hit the ceiling and found 4.8 CPU in untracked logic

The bot ate a steady 17.7 of 20 CPU, bucket sat around 2-22. Decomposition showed: 8.25 CPU is a hard intent ceiling, nothing to do. And 4.8 CPU is in untracked logic — that's where 90% of the potential lives. A story about module-level cache in an ephemeral runtime.

#screeps#performance#profiling#architecture#javascript

10 May 2026

MMO as production: Grafana, PostgreSQL and WebSocket for a game bot

Not every game lets you plug in your own monitoring. In Screeps the bot is plain JavaScript and the game world exposes WebSocket and REST. I built an external collector and watch the energy economy and global market like an exchange.

#screeps#observability#grafana#postgres#architecture

8 May 2026

Prompt is code: how I taught an LLM to stop inventing filters and to call SQL

I plugged a local LLM into OLT monitoring — and caught three categories of hallucinations. The fix wasn't a model upgrade, it was tool design and the system prompt.

#llm#prompt-engineering#mcp#tool-design

6 May 2026

SSE + Spring Security: three async-dispatch traps that break tool-streaming

I shipped SSE for tool-streaming an LLM and ran into three separate Spring Security bugs around async processing. Breakdown of each and the minimum correct code.

#spring-boot#spring-security#sse#java

4 May 2026

Local LLM in network monitoring: Ollama + MCP + Spring Boot in three days

How I plugged a local LLM into an OLT monitoring system through MCP. Architecture, an in-process dispatcher instead of HTTP, 16 tools, SSE to the browser.

#llm#mcp#spring-boot#ollama#case

2 May 2026

Aurora-mesh: A subtle background that doesn't turn your site into a SaaS landing page

How to add an atmospheric gradient background to an editorial portfolio without breaking its minimalist tone. Tradeoffs and the final masked-fade implementation.

#css#design#animation#editorial

2 May 2026

GetOLT: how a pile of telnet scripts grew into a PON-network monitoring service

Origin story of an OLT/ONU monitoring service in a telecom. From scattered scripts to a production tool that has been catching signal degradation and closing the 'subscriber → OLT/port/ONU' link for over a year.

#telecom#monitoring#pon#case#evolution