Monday’s applicant pile
A decent job post pulls in twenty or thirty applicants a week. Each one deserves a few minutes (the résumé, the application answers, the GitHub link that turns out either empty or astonishing), and the reading itself isn’t hard. Giving applicant twenty-six the attention applicant one got is the hard part, at four in the afternoon after back-to-back interviews, when a résumé with an odd layout gets skipped and nobody writes down that it happened.
Most ATS products have some screening built in by now, usually knockout questions or keyword matching. That handles “are you authorized to work in the US” well enough. It does much worse with anything that needs actual reading, like whether six years of “platform work” at a bank meant building services or babysitting a deploy script.
What dassi does in Greenhouse, Lever, or Ashby
It works through your ATS in your own browser, signed in as you, on the schedule you set. Workable and homegrown systems work the same way, since dassi reads the page rather than calling an API. For each new application it reads the résumé and the answers, scores them against your criteria, and leaves a note on the profile that quotes the résumé lines behind the score and lists what the application didn’t answer.
It doesn’t change stages. Shortlisted applicants get a phone-screen invitation drafted from your template and saved unsent, and the ones below your bar stay in whatever stage they were in, note attached, until you look.
Some applications it can’t call. A scanned PDF that comes through as garbage goes into an unclear pile instead of getting a guess, and so does the candidate with four years of backend work plus a two-year apprenticeship who’s applying against a five-year requirement. In the example run on this page two of twenty-six landed there. We’d expect more in the first week or two, while your scorecard still has soft spots in it.
Fairness depends on your criteria
A screening run can’t be fairer than the scorecard behind it, and we don’t think any tool can promise otherwise. What dassi adds is consistency and a paper trail. It applies your criteria to every applicant the same way and writes down its reasoning, so when a score looks off you can see why in about thirty seconds and overrule it. The decision on each candidate is yours.
That makes the criteria worth some care. Work authorization belongs on the list when the role requires it. Graduation years and a short list of approved schools usually don’t, because they tend to stand in for age and background. Rules on AI in hiring are also tightening in places (New York City’s Local Law 144 covers automated tools used in hiring decisions, and the EU AI Act treats recruiting uses of AI as high-risk), and whether they reach the way you use scores is worth asking an employment lawyer before you start.
Sourcing on LinkedIn and job boards
This part is optional. dassi can run your saved searches, read the profiles your account can already see at about the pace you’d read them, a dozen or so in a sitting, and log the promising ones to your hiring sheet with a link.
It sends no connection requests and no messages. LinkedIn caps how many invitations an account can send each week, and its user agreement restricts automated activity, so read those terms before you add LinkedIn to the job. dassi isn’t meant for mass invitations or scraping profiles at scale, and this workflow does neither. If you contact someone from the list, you write that message.
Before the first run
Runs happen in your browser, so it has to be open at 08:00. Résumé text and application answers go to the AI provider you chose so the model can read them, passing through dassi’s relay first if you’re on included credits. Check that your candidate privacy notice covers AI-assisted review.
Setup is a walkthrough. Screen three or four applicants while dassi watches and write a sentence on why each is in or out. Expect to edit the scorecard after the first run, mostly the phrases that meant something in your head and not much on paper, “strong communicator” being the usual suspect.