LinkedIn Optimizer

AI-powered LinkedIn growth system built by a non-developer Product Manager using Claude Code + Apify.

9 integrated skills that scrape LinkedIn data, benchmark your profile, generate content drafts, discover networking targets, and track your progress — all running locally with evidence-backed outputs.

What It Does

Skill	What It Does
Profile	Scores your profile on 10 dimensions against peer benchmarks
Content	Scrapes top PM/AI creators, generates 4-stage post drafts (Research → Hook → Body → Anti-slop)
Topics	Extracts trending topics from scraped posts, clusters by your content pillars
Jobs	Ranks PM jobs in your target region with fit scores, outreach drafts, referral targets
Network	Prioritizes connection targets by strategic value, generates personalized notes
Discover	Finds new people to follow from active post authors, enriches via Apify
Engage	Queues fresh posts with evidence-backed comment drafts
Watch	Competitor/peer signal digest from recent posts
Scorecard	Weekly operating view — score, stale sections, next action

Architecture

Dashboard (HTML) → POST /api/skills/{skill}/run → server.py
                                                      ↓
                                              SkillGateway
                                       ┌──────────┼──────────┐
                                   ApifyClient  LLMClient  ContentLab
                                       ↓           ↓          ↓
                                  LinkedIn      Claude     4-stage
                                  scraping      Haiku      pipeline
                                       ↓           ↓          ↓
                                   Scoring → Validators → Evidence Store
                                       ↓
                                  data/processed/{skill}.json
                                       ↓
                              Dashboard reads & renders

Key design decisions:

• Evidence-backed everything — every item has source, confidence, freshness, verified/unknown fields
• LLM can only draft from evidence — Claude generates text but cannot invent facts. Bad drafts are rejected and replaced with deterministic fallbacks
• Anti-slop validation — banned phrases ("game-changer", "leverage", "delighted to share"), placeholder enforcement, char limits
• Budget guard — live Apify usage from API, pre-flight exhaustion check, cost estimation per run

Tech Stack

• Python 3 (stdlib only — no pip dependencies except anthropic)
• Apify for LinkedIn scraping (profile, posts, jobs actors)
• Claude Haiku for evidence-bound drafting
• HTML/JS dashboard (no framework, vanilla)

Setup

# 1. Clone
git clone https://github.com/YOUR_USERNAME/linkedin-optimizer.git
cd linkedin-optimizer

# 2. Create .env (see .env.example)
cp .env.example .env
# Add your APIFY_TOKEN and ANTHROPIC_API_KEY

# 3. Create targets config
cp config/targets.example.json config/targets.json
# Edit with your LinkedIn URL, peers, and engagement targets

# 4. Install anthropic SDK
pip install anthropic

# 5. Run
./start.sh
# Opens http://localhost:8080/dashboard.html

Content Lab (4-Stage Draft Pipeline)

The content system doesn't just ask an LLM to "write a post." It runs a structured pipeline:

1. Research — Extracts key insight from inspiration posts, identifies your angle
2. Hook Writing — Generates 3 options (question, bold statement, story) — max 100 chars each
3. Body Draft — Full post with mandatory [INSERT YOUR REAL EXAMPLE] placeholders
4. Anti-slop Review — Validates against banned phrases, char limits, hashtag count. Auto-fixes and retries once.

If any stage fails, it falls back to deterministic output. The LLM is never trusted blindly.

Cost

Optimized for Apify Starter ($29/mo):

Skill	Posts/Profiles	Estimated Cost
Content	4 anchors × 3 posts	~$0.024
Engage	4 targets × 3 posts	~$0.024
Profile	1 profile	~$0.004
Jobs	25 results	~$0.025
Topics	Free (uses cached posts)	$0
Discover	Free (uses cached posts)	$0

Monthly estimate (3x content/week + 1x jobs/week + 1x profile/month): ~$3-4/month

Project Structure

linkedin-optimizer/
├── server.py                    # Local HTTP server, loads .env, routes to SkillGateway
├── start.sh                     # Launch script
├── dashboard.html               # 11-tab visual dashboard
├── config/
│   ├── settings.json            # Budget, actors, scoring weights
│   ├── content-pillars.json     # 4 content pillars + anti-slop rules
│   ├── job-filters.json         # PM job search filters
│   └── targets.example.json     # Template for your LinkedIn targets
├── linkedin_orchestrator/
│   ├── skills.py                # SkillGateway — 9 handlers, evidence-bound drafts
│   ├── apify_client.py          # Live budget sync, pre-flight checks, actor calls
│   ├── scoring.py               # Deterministic profile (10 dims), job (0-100), network scoring
│   ├── validators.py            # Anti-slop, evidence validation, relevance filtering
│   ├── content_lab.py           # 4-stage content pipeline
│   ├── topic_radar.py           # Topic extraction + pillar clustering
│   ├── discovery.py             # Rotating anchor discovery from post authors
│   └── common.py                # Evidence model, freshness, utilities
└── tests/
    └── test_linkedin_orchestrator.py  # 7 tests (scoring, validation, artifacts)

Built With

This entire system was built by a Product Manager with no prior coding experience, using Claude Code as the development partner. The orchestrator, dashboard, scoring engine, content pipeline, and all 9 skills were created through AI-assisted development — proof that PMs can build production tools, not just spec them.

License

MIT