LinkedIn Studio

The LinkedIn Studio at /admin/linkedin-studio generates copy-pasteable LinkedIn posts from your existing Arcturus-Prime content. It’s designed to turn technical blog posts, journal entries, and project updates into engagement-optimized LinkedIn content that showcases your engineering expertise without sounding salesy or AI-generated.

Why LinkedIn for Homelabs

The homelab and DevOps community is highly engaged on LinkedIn. Technical content that shows real problem-solving — debugging at 2am, architecture decisions, tool discoveries — performs well because it demonstrates hands-on expertise. LinkedIn’s algorithm now rewards dwell time (how long people read) over quick reactions, which favors detailed technical posts.

Four Input Modes

From Post

Browse all 76 blog posts and 73 journal entries, sorted by date. Select one and the AI generates a LinkedIn post from it. The full post content is loaded server-side and fed to the AI model — you don’t need to copy-paste anything.

Use the search bar to filter by title. Click a post to select it (highlighted with a blue border), then hit Generate.

From Scratch

Free-form text input for when you want to write about something that isn’t already a blog post. Enter a topic, paste notes, or describe what you want to share. The AI shapes it into LinkedIn format with proper hooks, structure, and hashtags.

Suggestions (Ideas)

Click “Scan for Posts” and the AI reviews your recent posts and journal entries to recommend the 5-8 best individual candidates for LinkedIn. Each suggestion includes:

Score — How LinkedIn-worthy the content is (0-100)
Suggested style — Which post format would work best
Reason — Why this would resonate on LinkedIn
Draft hook — A suggested opening line

Click a suggestion to select it, then Generate to create the post.

For multi-post narrative arcs instead of single posts, click “Plan a Series Instead” to switch to Journey mode.

Journey

AI-planned content series that looks at your entire blog and journal catalog to plan multi-post narrative arcs. Unlike Suggestions (which ranks individual posts), Journey mode thinks about the bigger story — what sequence of posts builds an audience, what threads connect multiple entries, what hasn’t been told yet.

Step 1: Plan — Click “Plan Content Series.” The AI receives metadata + 200-character excerpts from all entries plus your published history, and returns 5-8 post ideas. Each references 1-3 source posts to combine into one narrative, with a theme tag, suggested style, and narrative description.

Step 2: Generate — Click a plan item to select it (sets the source slugs and style), then hit Generate. The API loads the full body of the referenced sources (capped at 6K chars each) and generates a LinkedIn post that weaves them together.

The plan considers:

Chronological milestones — First server, first cluster, first automation
Thematic threads — Networking evolution, storage journey, build swarm arc
Origin story angles — Why homelab, what problem started it all
Achievements vs. struggles — Impressive numbers alongside relatable debugging
Published history — What stories have already been told (avoids repeats)

The Journey tab also shows a Published History sidebar at the bottom — the last 10 posts marked as published, with dates and LinkedIn URLs.

Five Post Styles

Style	Best For	Structure
Problem → Solution	Debugging stories, things you fixed	Hook → Context → Fix → Result → Takeaway → Question
Lessons Learned	Reflective insights from experience	Hook → What happened → 3 lessons → Reflection → Question
Tool Showcase	Specific tech you discovered or built	Hook → What it is → Why you chose it → Benefits → Limitation → Question
Architecture Decision	Trade-offs, tech choices	Hook → Options → Choice → Trade-offs → Hindsight → Question
Project Update	What’s new, what you shipped	Hook → Context → What you built → Why it matters → What’s next → Question

Tone Options

Casual — Natural, first-person, debugging-at-midnight voice (default, matches Arcturus-Prime style)
Professional — Slightly more polished, still authentic, good for employer visibility
Technical — Heavier on specifics, commands, and architecture details

Output Preview

The right panel shows your generated LinkedIn post with several preview sections:

Hook Preview

The first 210 characters are highlighted separately — this is the “See more” cutoff on LinkedIn. If the hook doesn’t grab attention in this space, people won’t click to read the rest.

Full Post

The complete post text with proper line breaks. LinkedIn posts render on mobile, so generous whitespace is important.

Character Count Bar

Visual indicator with markers at key thresholds:

Gray (< 1,300): Short — may underperform
Green (1,300-1,900): Sweet spot — best engagement
Yellow (1,900-2,500): Getting long but still fine
Red (> 2,500): Too long — consider shortening

Hashtags

3-5 relevant hashtags from the DevOps/homelab ecosystem: #homelab, #selfhosted, #devops, #linux, #docker, #kubernetes, #proxmox, #infrastructure, #automation, #networking, #sre.

First Comment

The Arcturus-Prime.com link and extra context go here, not in the main post. External links in the post body reduce LinkedIn reach by ~60%. Post the main text first, then immediately add this as your first comment.

Refine Buttons

After generating, five quick-action buttons let you iterate:

Shorten — Cut to under 1,500 characters, punchier
Expand — Add technical detail, specific numbers and commands
Better Hook — Rewrite the opening lines to be more scroll-stopping
More Human — Make it more conversational, less polished
Career Focus — Emphasize professional skills and expertise (good for employer visibility)

Amplify with Claude

Below the refine buttons, the Amplify with Claude button sends the current draft through Claude Sonnet 4.6 regardless of which model was used for the initial generation. This is the recommended workflow: draft cheap with a free model (Gemini Flash), then amplify with Claude for a quality pass that tightens language, sharpens the hook, and adds authenticity.

Copy to LinkedIn

Three buttons in the output panel header:

Copy post (clipboard icon) — Copies the full post text + hashtags
Copy first comment (comment icon) — Copies the first comment text
Mark as Published (bookmark icon) — Saves the post to history (see below)

Paste directly into LinkedIn. The line breaks are preserved.

Post History

After posting to LinkedIn, click the bookmark icon in the output header to mark it as published. You’ll be prompted for the optional LinkedIn post URL (paste it after posting). The post metadata is saved to data/linkedin-posts.json:

Full post text (for dedup reference)
Hashtags, source slugs, collections, style, model, character count
LinkedIn URL (if provided)
Theme tag

This history feeds back into Journey mode and Suggestions — the AI sees what’s been published and avoids suggesting the same topics. History is visible in the Journey tab sidebar showing the last 10 published posts with dates and links.

Identity Sanitization

All generated output is automatically run through the Galactic Identity System (identity_map.json, 148 patterns). Real hostnames, IPs, usernames, and infrastructure details are replaced with their fictional equivalents before display. You never need to manually sanitize LinkedIn posts.

LinkedIn Best Practices (Built In)

The AI prompts encode these research-backed practices:

No external links in body — Put links in first comment
Hook-first writing — Bold statement or vulnerability in first 2 lines
Line breaks for mobile — Dense text gets skipped on phones
Genuine questions — Not engagement bait, real discussion starters
Specifics over generics — Exact numbers, tech names, real outcomes
3-5 hashtags max — More reduces reach

Model Selector

The top bar model selector offers 9 models across 4 price tiers:

Tier	Models	Cost
Free	Gemini 2.0 Flash, Llama 3.3 70B, DeepSeek R1	$0
Budget	Gemini 2.5 Flash, DeepSeek V3	~$0.15-0.30/M tokens
Mid	Claude Haiku 4.5, Gemini 2.5 Pro	~$1-2/M tokens
Premium	Claude Sonnet 4.6, Claude Opus 4.6	~$3-5/M tokens

Provider-aware routing sends requests directly to Anthropic, Google, Groq, or NVIDIA APIs when keys are configured, bypassing OpenRouter’s markup. Falls back to OpenRouter for any model.

The Amplify with Claude button always uses Claude Sonnet regardless of the selected model — it’s a quality pass on top of a cheaper draft.

API

POST /api/admin/linkedin-post with these actions:

generate — Create a LinkedIn post from content, slug, or freeform input
suggest — AI scans recent content for LinkedIn-worthy candidates
refine — Iterate on a generated post with specific instructions
journey — Plan a multi-post content series from the full content catalog
journey-generate — Generate a post from multiple source entries with a narrative angle
save-published — Save a post to the published history tracker
get-history — Load the published post history

Content Lab (/admin/content-lab) — Full blog post and journal generation
Probe Studio (/admin/probe-studio) — Content analysis and coaching
Voice Engine (src/lib/voice-engine.ts) — Shared AI calling infrastructure