I'm an ML engineer, and job hunting felt like a second job. You search, open a posting, read it, work out whether you fit, and repeat. Oink is what I built to do that first pass: it turns your current brief into explained matches, delivered through Telegram or the Oink iOS app. It is built and operated alongside the nomad-agent collection tools published on Apify. This site documents the product, the reasoning behind its matching, and practical ways to look at public job listings more carefully.
I write these pages to help you make one concrete decision. I don't write them to cover every keyword variation, and Oink does not publish copied job descriptions as landing pages.
How Oink creates and checks content
Before a page goes live, I check its product claims against the current public interface, the configuration, and the code that actually runs. Technical claims about third parties should link to the organization's own documentation or another primary source. If a finding comes from my own data, the page says the sample window, the method, and the limits, and does not present one anecdote as a rule.
AI-assisted drafting and organization may be used. That doesn't make a claim true, and I don't treat it as a source. I stay responsible for checking how the product behaves, resolving contradictions, and deciding whether a page is ready. This disclosure follows the "who, how, and why" questions in Google's people-first content guidance.
Each evergreen page shows a Last reviewed date. Articles show a publication date and, after a material change, a separate updated date. Rebuilding the site doesn't move either date, and the sitemap uses the same date, so a deployment never pretends unchanged content is new.
Four kinds of proof
Oink keeps these apart, because they are easy to confuse:
- A local test shows that a defined behavior passed in the tested tree.
- A deployment receipt identifies what was uploaded and where.
- A live fetch shows what a public URL returned at one moment.
- A sitemap or IndexNow ping is a request to look, not proof of indexing. An indexed URL is not proof of impressions, clicks or a useful conversion.
The same rule applies to the product. A Telegram card that was sent doesn't prove that anyone opened it or applied. The 0 to 5 match score ranks relevance against your current brief, and it is not an interview probability. A liveness check lowers the risk of a stale posting, but it can't guarantee that an employer still accepts applications. The AI matching guide and the job-source guide cover those limits in detail.
Commercial relationships and links
Some developer articles link to Apify Actors published by nomad-agent, the same operator behind Oink's collection work. Those links promote related tools of mine. They are not independent reviews. Actor availability, inputs, pricing and source support change, so check the current Store listing and run contract before you integrate.
If you're building job alerts, trackers or career tools, the related Job Atlas project offers job-data tools and integration guides. Oink is the product for job seekers.
Calls to action for job seekers open Oink itself. A guide should stay useful without a click on them, and a promotional link is never a reason to include an unsupported statistic or hide a limit.
Privacy and responsible reporting
The public privacy policy describes the account, resume, preference, interaction, diagnostic and analytics data the product can process. Open-web discovery can send job-title or keyword queries derived from your profile to a search provider. It doesn't need your raw resume for that request. Read the current Privacy page for the actual disclosure, and don't assume "AI job search" means no third-party processing.
When I publish research from our own data, it is aggregate. It must not expose resumes, profile text, contact details, chat identifiers or anything that identifies a job seeker. Small groups are suppressed or combined when reporting them could identify someone.
Corrections and material changes
When I find a factual problem, I check the page against the authoritative source and correct it narrowly. The updated date changes only if the content you read changed in a meaningful way. If the evidence is missing or contradictory, I qualify the claim or remove it, and I don't polish it into certainty.
Product facts drift too: source availability, pricing, onboarding steps and delivery behavior all change. Older articles get reviewed so stale counts, outdated score wording and claims stronger than the product can prove come out.
To understand the product rather than how I publish, start with how Oink works. For practical reading, browse the job-seeker and developer guides, or see how a match becomes a Telegram job alert.