🎓 LinkedIn Profile — Courses & Recommendations Signal

Scrapes public LinkedIn profiles and turns two fields most LinkedIn tools ignore — the Recommended-courses module and received Recommendations — into a lightweight, transparent signal layer. Same profile scrape as the base LinkedIn Profile Scraper, plus course_signal and recommendation_signal per profile, plus a run-level signal_summary.

What it does

Two input modes:

• profiles (default, reliable): scrape the exact profile URLs you provide.
• company: discover a company's employees (company pages + Google search), then scrape them. Google discovery is brittle and is usually blocked from datacenter IPs — see Limitations.

For each scraped profile it emits every base field, then adds:

• course_signal — course_count, total_learning_minutes (sum of parsed scraped course durations), course_titles (passthrough), themes, and matched_keywords.
• recommendation_signal — count, recommenders (passthrough of scraped authors), themes, and matched_keywords.
• ai (optional, opt-in) — seniority, skills, lead_quality, rationale from an LLM over the profile's free text.
• scrapedAt — absolute ISO timestamp.

A run-level signal_summary (counts and theme frequencies) is written to the run's key-value store.

Themes are real, not fabricated

A theme is emitted only when a keyword literally begins a word in the scraped text (\b word-start match). This stops ship matching leadership, api matching capital, and git matching digital, while still catching intended stems (data → database, innovat → innovative). Every theme ships with its matched_keywords for provenance. No score, rating, or constant produces an output number.

Input

Field	Type	Notes
mode	enum profiles | company	Default profiles.
profileUrls	string list	Used in profiles mode. Full URLs or bare slugs.
companies	string list	Used in company mode. Slugs or company URLs.
maxEmployees	integer	company mode cap per company. Default 20.
rawData	boolean	Emit plain base records only (no signals). Default off.
aiEnhancement	boolean	Enable the LLM ai block. Default off.
aiModel	enum	Provider auto-detected from name (Anthropic / OpenAI / Google / xAI / DeepSeek / Perplexity / Mistral).
aiApiKey	secret	Falls back to the provider env var if blank.
proxyConfiguration	proxy	Apify Proxy, RESIDENTIAL strongly recommended.

Output (dataset table columns)

public_identifier, fullname, headline, current_company, location, course_count, courseThemes, recommendation_count, recThemes, profile_url, scrapedAt. Full nested course_signal / recommendation_signal blocks are in each JSON row.

Limitations (honest)

• Anti-bot / authwall. LinkedIn blocks datacenter IPs with 999 / authwall. Use a residential proxy. Blocked profiles come back as {profile_url, error} rows — never fabricated data.
• Google employee discovery (company mode) is usually blocked from cloud IPs and may return few or zero URLs. It never invents profile URLs. profiles mode is the reliable path.
• is_creator / is_influencer / is_premium / show_follower_count are not scraped from the public profile and are emitted as null (the base hardcoded them to False/True; that was corrected here).

Keywords: linkedin scraper, linkedin profile scraper, linkedin recommendations, linkedin learning courses, lead enrichment, sales prospecting, people data, profile signals, skills classification, seniority detection, b2b leads, apify.