Day 19 opened with a clean site — zero orphans, zero broken links confirmed
by check_ai_tree.sh -a before the session began. Three new content
pages were created: a Prompt Engineering guide for SysAdmins in learning/,
and two new CLI tool deep-dives (jq and ollama) in
deep-dive/. The learning-index.html was updated to add
the Prompt Engineering card. The deep-dive-index.html update is
deferred to Day 20 when fzf.html and gh.html will also
be ready — all four new cards will be added in one pass.
check_ai_tree.sh -a — clean, zero issuesdeep-dive/index.html was a full copy, not a symlink —
corrected: rm index.html && ln -s deep-dive-index.html index.htmllearning/prompt-engineering-for-sysadmins.html —
Comprehensive prompting guide drawn from 19 days of real daily use. Covers
context, output format specification, roles, iterative refinement, session
management, verification habits, and a quick-reference pattern table.
Includes before/after prompt comparisons and Craig's personal take.deep-dive/jq.html — JSON processing at the command line.
Covers installation, basic filters, array operations, AI-assisted filter
generation workflow, real SysAdmin examples (Docker, journald, AWS CLI,
Kubernetes), useful patterns, and strengths/limitations.deep-dive/ollama.html — Running local LLMs on Linux.
Covers installation as a systemd service, model selection table, basic usage,
REST API, integration with sgpt/aider/fabric, multi-user network setup,
security considerations, and the case for local vs cloud AI.learning/learning-index.html — Prompt Engineering card
added to Episode 3 section, positioned after Introduction to AI and before
How AI Actually Works. Footer date updated to March 16, 2026.deep-dive/fzf.html — fuzzy finder with AI integrationdeep-dive/gh.html — GitHub CLI deep-divedeep-dive/deep-dive-index.html — update with all four
new cards (jq, ollama, fzf, gh) in one pass after all pages are completeThe Prompt Engineering page is the most directly personal page on the site — drawn entirely from real experience rather than documentation. The before/after prompt comparison format worked well visually and makes the advice concrete rather than generic. This page will age well as the project continues because the nuances it documents are earned, not theoretical.
The jq + AI combination is genuinely one of the most practical topics in the series. jq is essential but difficult; AI makes it accessible. The pattern of "describe what you want, get a filter, test it, refine" is a workflow that delivers immediate value.
Ollama completes a gap that was referenced in three earlier pages (sgpt, aider, fabric) but never documented on its own. The multi-user LAN inference server use case is particularly relevant for institutional environments like BU.
Decision to batch the deep-dive index update with Day 20 content was correct — one clean update with all four new cards is better than two partial updates.
cmh — AI Education Project Day 19 — March 16, 2026
Next session: Day 20 — fzf.html, gh.html, deep-dive-index.html update