I run freehire, an open-source job aggregator that ingests postings from dozens of ATS platforms (Greenhouse, Lever, Ashby, Workday, etc.), normalizes them into one schema, and runs each through a facet pipeline.<\/p>\n
Just exported the whole catalogue to Hugging Face: freehire-jobs \u2014 6,041,471 postings.<\/p>\n
Each row is a raw posting (title, company, description, URL, source, posted_at) plus derived facets:<\/p>\n
– Dictionary-only (deterministic, no LLM): skills, seniority, category, work_mode, posting_language, employment_type, education_level, english_level, experience_years_min<\/p>\n
– Dictionary-first, LLM-filled: countries, regions, cities<\/p>\n
– LLM-only: salary_min\/max, currency, period \u2014 pay is stated in every format imaginable, no dictionary handles that<\/p>\n
Company rows carry their own facets too (industries, HQ country, size, YC batch\/stage where applicable).<\/p>\n
Format: 20 gzip JSONL shards, split by row range for easy streaming\/resuming.<\/p>\n
Backstory: I originally wanted to train a cheap classifier for seniority\/category tagging instead of paying for an LLM call per job. Turned out the dictionary+LLM pipeline already in prod beats a from-scratch classifier by enough that finishing the classifier wasn’t worth it \u2014 so I’m sharing the data instead.<\/p>\n<\/div>\n
submitted by \/u\/Dry-Library-8484 <\/a>
[link]<\/a><\/span> [comments]<\/a><\/span><\/p>
