> Where music, humanity, and technology intersect
From blank directory to a scaffolded autonomous agent project — designing an email-triggered AI harness on a Hostinger VPS using the Ralph Wiggum methodology, with a path from Claude API to local Qwen3 models.
Read More →Starting from a lab spec about city emergency coordination, we designed a UML class diagram in draw.io, mapped it to a SQLite schema, and built a working menu-driven CLI that demonstrates incident reporting, unit dispatch, and the availability guard.
Read More →A slow, deliberate walkthrough of four graduate DS/ML exercises — vectorized confusion matrices, Python descriptors, context managers, and lazy generator pipelines — with the emphasis on understanding over completion.
Read More →A session spent red-teaming a set of Python security utilities — finding bypass vectors in sensitive field masking, then implementing hardened replacements with 62 passing tests.
Read More →How I built a 10-module Elixir learning repository with detailed READMEs, exercises, and a blog draft — then published it to GitHub in one session.
Read More →A Ralph Docker session ran overnight with unlimited iterations, creating 67 empty workspace branches and consuming 71% of my weekly Sonnet token allotment — with zero productive output. Here's the forensic analysis and the fix.
Read More →Ralph-docker now has a built-in setup command that interviews you and generates all project files via Claude — no external skill required. Plus, a rewrite of the README to put observability front and center.
Read More →A debugging session where fixing one CSS overflow led to discovering Hugo's hidden table-based syntax highlighting, duplicate asset files, and the joys of browser caching.
Read More →A session where Claude Code reviewed four open PRs on a Go/RabbitMQ event system, identified bugs ranging from mutex contention to JSON schema mismatches, and fixed them all — merging three and leaving one ready to go.
Read More →Building a multi-agent system incrementally across 4 assignments — agent loops, pluggable tools, bounded memory with escalation, and orchestration — all in pure Python with 50 tests and zero LLM dependencies.
Read More →