AI & Data Jobs Finder | HiringSignal

Turn selected public employer job boards into a shortlist for your AI/Data newsletter or niche job board. Choose a curated company group and role preset, then download matching jobs as CSV, JSON, RSS, JSON Feed and a Markdown newsletter draft. Greenhouse, Lever and Ashby feeds share one normalized schema with source links.

Step-by-step first run · Hướng dẫn tiếng Việt

Start with six matching jobs

{
  "companyGroup": "ai",
  "rolePreset": "data-ai",
  "maxJobs": 6,
  "maxJobsPerBoard": 2
}

Run this small input, inspect the jobs and their application requirements, then open Output → Newsletter jobs shortlist (Markdown) for a draft you can edit. This input saves at most six matching jobs and at most two per board; availability changes, so it may return fewer or zero. Six saved rows correspond to $0.009 in job-saved events at the configured price. Check Pricing and the run's platform charges.

The watchlist is a starting point, not a complete AI-company directory, a global job search or an endorsement by listed employers. The newsletter contains only saved rows in provider order. It is not a newest-first ranking and is not posted anywhere automatically.

Pick your employer watchlist

companyGroup accepts none, ai, developer-tools, fintech or all. The built-in catalog associates selected employer display names with public ATS board keys. Inspect the current EMPLOYER_CATALOG output or run the standalone CLI's hiringsignal catalog command for the actual roster and preset terms. See the catalog notes for scope and verification context.

You can also supply boards:

{
  "companyGroup": "none",
  "rolePreset": "engineering",
  "boards": ["greenhouse:stripe", "ashby:ashby"],
  "maxJobs": 6,
  "maxJobsPerBoard": 3,
  "knownJobIds": []
}

When both are supplied, explicit boards run first and the company group's boards are added without duplicates. Use companyGroup: "none" for only your own boards. The combined selection must contain 1–30 unique boards. To try only the AI preset, clear any older explicit boards input.

Supported board forms include greenhouse:stripe, lever:leverdemo, lever-eu:yourboard, ashby:ashby, and HTTPS public board URLs on boards.greenhouse.io, job-boards.greenhouse.io, jobs.lever.co, jobs.eu.lever.co and jobs.ashbyhq.com. Ashby capitalization is preserved. Arbitrary API endpoints, custom careers domains and redirects are not accepted.

HiringSignal reads official public JSON feeds. It needs no ATS API key or proxy. It does not submit applications, read private candidate data, bypass access restrictions or support LinkedIn, Indeed or social-media scraping.

Refine matching jobs

rolePreset accepts all, data-ai, engineering, product-design or sales-marketing. Presets match case-insensitive title words or phrases from the catalog; they are practical title filters, not an AI assessment of the role. The all preset adds no role restriction.

Optional includeTitleKeywords, excludeTitleKeywords and locationKeywords use case-insensitive substring matching. Terms within each list are OR; different filter lists and the role preset combine with AND. Exclusion takes priority. All filters apply before saving and billing.

employmentTypes is an optional OR list of full-time, part-time, contract, temporary, internship, freelance, apprenticeship and unknown. An empty list adds no restriction. employmentTypeNormalized maps only unambiguous explicit provider labels such as Lever categories.commitment and Ashby employmentType; the original label remains in employmentType and provenance in fieldSources. Missing/ambiguous labels stay unknown. Greenhouse jobs usually have no explicit employment type and are excluded when you select only known types. The title and description never determine employment type.

remoteOnly requires an explicit provider remote field, excluding hybrid and unknown jobs. A remote-looking title or location is not enough. Greenhouse's list feed generally has no explicit workplace field, so its jobs usually do not pass this filter. Remote does not mean eligible to work from every country. Check each source posting's location, work authorization and application terms.

publishedAfter is an inclusive ISO date or timestamp. When set, jobs with an unknown publication date are excluded. Greenhouse updated_at is a modification date and never substitutes for publication. Lever public jobs generally lack a documented publication date. Ashby publishedAt means last published, not necessarily first posted. publishedAtMeaning preserves that distinction.

Defaults companyGroup: "none" and rolePreset: "all" preserve existing manual watchlist workflows. The displayed starter input chooses AI/Data explicitly.

Output you can use

The dataset has one row per normalized, filtered job actually saved:

Fields	Meaning
id, jobKey, source, boardKey	Source ID and stable provider/board/job identity
company, companySource	Employer board slug and its source
companyDisplayName, companyDisplayNameSource	Curated display label when known, otherwise the board slug, with provenance
title, locations, department, team	Published job fields when available
workplaceType, workplaceTypeSource	Explicit remote/hybrid/onsite value or unknown
employmentType, employmentTypeNormalized, descriptionText	Original employment label, exact normalized label or unknown, and cleaned job content
jobUrl, applyUrl	Source job/application links when exposed
publishedAt, publishedAtMeaning, updatedAt	Publication and modification dates kept distinct
salary	Source compensation only; null when absent
fieldSources, requestUrl, observedAt	JSON field mapping, source API and observation time

The Output tab also links spreadsheet-safe CSV, newsletter.md, SUMMARY, DIAGNOSTICS, KNOWN_JOB_IDS, EMPLOYER_CATALOG and complete OUTPUT JSON. The Markdown draft includes titles, employer labels, locations, workplace labels and source application/job links. Review it before republishing.

jobs.rss (RSS 2.0) and jobs.feed.json (JSON Feed 1.1) export only the saved rows, in provider order, with stable jobKey item IDs and source job links. Missing source publication dates are omitted; observation time never becomes a publication date. Ashby's last-publication meaning is labeled in the content. Feed descriptions are shortened to 2,000 source description characters; open the posting or use dataset JSON for more content. RSS source markup is escaped and JSON Feed uses plain text. These are static per-run exports, not an automatically updated hosted subscription or an unlimited job archive. See the integration guide.

Diagnostics report per-board fetched/filtered/invalid/saved counts, errors and completion flags. Summary signals use only saved rows, not total employer hiring. A board name in the catalog does not guarantee its feed is currently available or contains matching roles.

If every attempted employer feed fails, the run reports failure with sourceFailureFatal: true after writing diagnostics, empty exports and unchanged known IDs. A nonempty feed in which no listed job can be normalized is a source schema failure. Successful empty feeds and zero filter matches are valid results; mixed failures and successes remain partial and are visible in diagnostics.

Repeat without paying for known keys

Copy the KNOWN_JOB_IDS JSON array into the next run's knownJobIds. Only previous keys and jobs actually saved in this run enter this rolling union. Filtered, failed or unpaid jobs are not marked seen. Keys are checked before saving and billing.

Up to 10,000 keys are retained; the oldest are dropped and knownJobIdsTruncated is true if the union grows larger. An evicted key can appear again. This is a bounded seen-list, not an unlimited history database. Scheduling the same input does not automatically copy state: your workflow must save the new array and supply it in the following run.

An unknown key means not in your supplied seen-list, not newly published. Previously saved jobs with changed descriptions remain skipped by key. An absent job is never labeled closed. Capped, filtered, rate-limited or failed boards cannot establish a full employer snapshot or hiring growth.

The standalone CLI can load a previous export directly:

hiringsignal run --input examples/ai-data-newsletter.json --previous-ids output/known_job_ids.json --output output-next

The combined input seen-list must still fit the 10,000-key bound.

Pricing and limits

The configured Actor fee is $0.0015 per saved job ($1.50 per 1,000). The Store Pricing tab controls current terms. In pay-per-event mode, job-saved occurs once per successfully normalized, filtered, saved dataset row. Errors, empty boards, filtered jobs and known keys do not generate this event. Check displayed platform charges as well as the Actor price; infrastructure may still be used when no row is saved. Your spending limit bounds persisted results.

Limits: 30 selected boards, 1,000 saved jobs per run, 1,000 per board, 8 MiB decoded JSON per response, 80 HTTP attempts per run, 20 Lever pages per board, 5,000 processed source records per board and 12,000 description characters per saved job. Transient requests receive at most one retry with paced requests and a bounded retry wait. No JavaScript pages or private ATS endpoints are read.

Source fields may be absent or change. Salary is not estimated; currencies and compensation intervals remain as published. Check source links and diagnostics before making hiring or research decisions. Respect source content terms when republishing jobs.

Support

Open the Actor's Issues tab with the provider, public board, run ID, expected field and observed result. Remove tokens and private data from reports.