{"id":41393,"date":"2026-06-17T15:27:15","date_gmt":"2026-06-17T13:27:15","guid":{"rendered":"https:\/\/www.graviton.at\/letterswaplibrary\/self-promotion-paid-built-a-deterministic-job-postings-data-pipeline-looking-for-feedback\/"},"modified":"2026-06-17T15:27:15","modified_gmt":"2026-06-17T13:27:15","slug":"self-promotion-paid-built-a-deterministic-job-postings-data-pipeline-looking-for-feedback","status":"publish","type":"post","link":"https:\/\/www.graviton.at\/letterswaplibrary\/self-promotion-paid-built-a-deterministic-job-postings-data-pipeline-looking-for-feedback\/","title":{"rendered":"[self-promotion] [PAID] Built A Deterministic Job Postings Data Pipeline: Looking For Feedback"},"content":{"rendered":"<p><!-- SC_OFF --><\/p>\n<div class=\"md\">\n<p><strong>Disclosure:<\/strong> I built this project and this is my own API\/product. It has free and paid access tiers. I\u2019m sharing it here because I think the data engineering approach may be useful, and I\u2019m looking for technical feedback.<\/p>\n<p>I built Trace Jobs Core, a job postings data API built around a simple idea: <strong>Do not guess.<\/strong><\/p>\n<p>A lot of job data pipelines end up doing some combination of:<\/p>\n<ul>\n<li>scraping HTML pages<\/li>\n<li>parsing unstable frontend output<\/li>\n<li>using models to extract fields<\/li>\n<li>guessing missing\/ambiguous values<\/li>\n<li>deduplicating after the fact<\/li>\n<\/ul>\n<p>I took a different approach.<\/p>\n<p>The pipeline ingests job postings from public machine-readable sources, translates them into a <a href=\"http:\/\/schema.org\/\">Schema.org<\/a> JobPosting format, applies only deterministic normalization where the source provides clear structure, and preserves original values when fields are ambiguous.<\/p>\n<p>Current system:<\/p>\n<ul>\n<li>9,800+ structured feeds<\/li>\n<li>~13k new postings\/day<\/li>\n<li>daily refresh<\/li>\n<li><a href=\"http:\/\/schema.org\/\">Schema.org<\/a> JobPosting records<\/li>\n<li>SHA-256 based deduplication<\/li>\n<li>RFC 8785 canonicalization<\/li>\n<li>original upstream values preserved when normalization is uncertain<\/li>\n<\/ul>\n<p>The goal is not to create a &#8220;smart&#8221; interpretation layer. The goal is to provide stable, predictable data and leave interpretation to the downstream user. <\/p>\n<p>A future enrichment layer could exist separately, but it would remain separate from the source-faithful data layer.<\/p>\n<p>Examples (HTML + JSON responses refreshed daily):<br \/> <a href=\"https:\/\/kaleh.net\/trace\/examples.html\">https:\/\/kaleh.net\/trace\/examples.html<\/a><\/p>\n<p>Documentation:<br \/> <a href=\"https:\/\/kaleh.net\/trace\/docs.html\">https:\/\/kaleh.net\/trace\/docs.html<\/a><\/p>\n<p>Project overview:<br \/> <a href=\"https:\/\/kaleh.net\/trace\/\">https:\/\/kaleh.net\/trace\/<\/a><\/p>\n<p>I would especially appreciate feedback on:<\/p>\n<ul>\n<li>dataset design<\/li>\n<li>normalization strategies<\/li>\n<li>preserving source fidelity<\/li>\n<li>handling schema differences between providers<\/li>\n<li>what fields\/data would make this more useful<\/li>\n<\/ul>\n<p>Thanks!<\/p>\n<\/div>\n<p><!-- SC_ON -->   submitted by   <a href=\"https:\/\/www.reddit.com\/user\/0o3705\"> \/u\/0o3705 <\/a> <br \/> <span><a href=\"https:\/\/www.reddit.com\/r\/datasets\/comments\/1u89cef\/selfpromotion_paid_built_a_deterministic_job\/\">[link]<\/a><\/span>   <span><a href=\"https:\/\/www.reddit.com\/r\/datasets\/comments\/1u89cef\/selfpromotion_paid_built_a_deterministic_job\/\">[comments]<\/a><\/span><\/p><div class='watch-action'><div class='watch-position align-right'><div class='action-like'><a class='lbg-style1 like-41393 jlk' href='javascript:void(0)' data-task='like' data-post_id='41393' data-nonce='bc39e8310e' rel='nofollow'><img class='wti-pixel' src='https:\/\/www.graviton.at\/letterswaplibrary\/wp-content\/plugins\/wti-like-post\/images\/pixel.gif' title='Like' \/><span class='lc-41393 lc'>0<\/span><\/a><\/div><\/div> <div class='status-41393 status align-right'><\/div><\/div><div class='wti-clear'><\/div>","protected":false},"excerpt":{"rendered":"<p>Disclosure: I built this project and this is my own API\/product. It has free and paid access&#8230;<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[85],"tags":[],"class_list":["post-41393","post","type-post","status-publish","format-standard","hentry","category-datatards","wpcat-85-id"],"_links":{"self":[{"href":"https:\/\/www.graviton.at\/letterswaplibrary\/wp-json\/wp\/v2\/posts\/41393","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.graviton.at\/letterswaplibrary\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.graviton.at\/letterswaplibrary\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.graviton.at\/letterswaplibrary\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.graviton.at\/letterswaplibrary\/wp-json\/wp\/v2\/comments?post=41393"}],"version-history":[{"count":0,"href":"https:\/\/www.graviton.at\/letterswaplibrary\/wp-json\/wp\/v2\/posts\/41393\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.graviton.at\/letterswaplibrary\/wp-json\/wp\/v2\/media?parent=41393"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.graviton.at\/letterswaplibrary\/wp-json\/wp\/v2\/categories?post=41393"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.graviton.at\/letterswaplibrary\/wp-json\/wp\/v2\/tags?post=41393"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}