uses

What’s on the machine when I’m working, May 2026. Updated occasionally; if anything below looks stale, it probably is.

hardware

• dev: M-series Mac with Cursor as the day-job editor, VS Code and Sublime kept warm for the things they’re each best at, iTerm2 with MesloLGS Nerd Font as the terminal.
• maral: a spare 2023 MacBook Air M2 16GB on the LAN running a local LLM stack — Ollama serving qwen3:14b to Claude Code over ANTHROPIC_BASE_URL. Full story.
• pi 5: a Raspberry Pi 5 16GB with a Hailo-10H AI HAT+ 2. Was meant to be the LLM appliance; turned out to be a great vision / Whisper / experimentation box and a slow agent backend.
• router: a Google WiFi mesh of six AC-1304 pucks running OpenWRT, unbricked the hard way.

shell + system

• zsh + oh-my-zsh, with bounded marker blocks in ~/.zshrc for project-specific functions so they don’t bleed into each other.
• macOS 26.3.
• Porkbun for DNS (this domain). ProtonMail for inbox.

ai / llm

• Claude Code is the primary work driver, used both directly and through a router that defaults to the local Maral and requires explicit claude-cloud to spend cloud quota.
• MCP for tool plumbing — at work, an internal MCP server + Claude-skill suite drives Jira, Bitbucket, and AWS Cost Explorer.
• AWS Bedrock + Anthropic API day-job stack: Knowledge Bases for RAG, Lambda for orchestration, pgvector and OpenSearch k-NN for vector search.
• qwen3 family for the local side. qwen3:14b on Maral, qwen3:8b as the small-fast fallback.

languages + frameworks (at work)

• TypeScript + React 19 + Next.js App Router + Tailwind 4 on the frontend.
• Node 20+ or Python 3.12 on the backend, depending on the service.
• Scala / Akka / Play if you go far enough back — that was the Flow Commerce era. Not the daily driver anymore.
• Prisma as the ORM when there’s a DB.
• Postgres for OLTP, RDS with a Glue/Python mirror of partner Redshift bi schemas for analytics.

infra

• AWS ECS Fargate + ALB + ECR for long-running services.
• AWS Lambda (container image) + EventBridge Scheduler for headless jobs.
• AWS Glue (Python) for batch / pipeline work.
• Terraform for all of the above, S3 + DynamoDB backend.
• Bitbucket Pipelines for CI/CD.
• Secrets Manager for secrets, SSM Parameter Store for non-secret config.

the blog itself

• Jekyll, github-pages gem, kramdown + rouge for syntax highlighting.
• jekyll-seo-tag for OG / meta. Per-post 1200x630 cards rendered by a small Playwright script in _scripts/og.py.
• Hosted on GitHub Pages with a Let’s Encrypt cert for ollyconn.com, fronted by Porkbun DNS.