MMOOZZGG.com

How to build an AI agent

AI recruiting agent: screening applications and first interviews

5 min read6 sources

An AI recruiting agent reads every application, asks candidates the first questions and books suitable people for an interview. A person keeps the hiring decision: under 152-FZ, Russia’s personal data law, decisions with legal effect based only on automated processing are banned without the candidate’s written consent.

On this page · 4 sections

A developer is needed

Screening applications

The agent takes applications from a job site, compares each resume with the requirements and gives a score with a reason. Instead of a hundred unread resumes, the recruiter gets a short list with an explanation for each candidate.

Through the API of hh.ru, a Russian job board, an employer can get the list of applications, change their status and invite a candidate to a vacancy. A developer does the connection: the app is registered on dev.hh.ru and given employer access.

The manager writes the criteria, not the agent. Experience, skills, schedule, city: each requirement can be checked, and each has a weight. The agent only applies them the same way to everyone.

The score has two parts: hard requirements, without which a candidate does not fit, and preferred ones, which move a candidate up the list. A candidate who fails a hard requirement does not disappear: they go to a separate list that the recruiter reviews once a week.

A model repeats the bias of the data it learned from. In 2018 it became known that Amazon’s resume screening tool, trained on resumes of current engineers, mostly men, ranked down resumes with the word “women’s”. Amazon stopped using it.

In Russia, such bias breaks the law. Article 64 of the Labor Code bans unjustified refusal to hire and limits based on sex, race, nationality, language, origin, property and family status. So sex, photo and marital status are removed from the agent’s data, and the recruiter spot-checks its scores.

First questions

After screening, the agent sends the candidate 3–5 questions about the vacancy: experience with the required software, readiness for the schedule, salary expectations. The answers go into the candidate card next to the resume score.

Questions are short, one at a time, in plain words. Candidates answer from their phones, and a twenty-question form loses people who would have answered three.

Choose the channel by where candidates reply. In the hh.ru API, the methods for messaging on behalf of the employer’s manager are outdated, so the agent asks its questions in Telegram, another messenger or through a question form.

The agent does not reject anyone on its own. Article 16 of 152-FZ bans decisions with legal effect based only on automated processing of personal data, except with the person’s written consent or in cases the law names directly.

If such consent exists, the operator must explain to the candidate how the decision was made, give a chance to object and review the objection within 30 days. It is simpler and fairer to leave the decision to the recruiter and the preparation to the agent.

Under Article 64 of the Labor Code, a candidate may demand the reason for a refusal in writing, and the employer answers in writing within 7 working days. The agent’s score with reasons helps the recruiter answer to the point and on time.

Booking the interview

The agent offers a suitable candidate free slots from the calendar of the recruiter or the manager, books the chosen time and sends a reminder the day before.

The agent handles rescheduling and cancellation the same way: the candidate writes that they cannot make it and gets new slots. The recruiter sees the result in the calendar, not the chat.

The candidate’s status changes in the system with no manual work. Through the hh.ru API, the employer invites the candidate to the vacancy, and a meeting with the candidate’s answers appears in the CRM or a spreadsheet.

If the candidate does not reply, the agent writes again after a day and then closes the application with a polite answer. Every candidate gets a reply, and that is the employer’s reputation.

Integration with a job site, a calendar and a CRM needs a developer. Our pilot on one vacancy costs from $1,499 and takes 14 days, with a metric in the contract: the share of screened applications and the time to the first reply to a candidate.

The page on AI for recruiting describes how we build such an agent and for whom it pays off.

Candidates’ personal data

A resume is personal data: name, phone number, experience, employers, sometimes a photo. A hiring agent works with it from the first application.

The candidate gives consent to data processing before the first questions, in the application or in the agent’s first message. The company’s lawyer prepares the consent text, and the agent only shows it and saves the answer.

Under 152-FZ, the personal data of Russian citizens is collected and stored in databases in Russia. If the agent sends a resume to a foreign model, that is a cross-border transfer, and before it the operator files a notice with Roskomnadzor, the Russian regulator.

A mistake is expensive. Since 30 May 2025, a company pays 1 to 3% of revenue for a repeated personal data leak, but at least $300,000 and at most $6 million.

So the hiring agent runs on a model and a server in Russia, the CRM stores only what the vacancy needs, and the recruiter and the agent see resumes, but not all employees do. The page on data security shows how we set this up in projects.

The agent’s rights, its action log and the checks before launch are covered in the AI agent security checklist.

Sources and fact-checking

Checked by the editors on 16 October 2026.

  1. Brace Law Firm, 9 July 2023 (brace-lf.com)
  2. ppt.ru, 30 November 2024 (ppt.ru)
  3. ACLU, 12 October 2018 (aclu.org)
  4. Labor Code of the Russian Federation, Article 64, legalacts.ru, 30 December 2001 (legalacts.ru)
  5. ConsultantPlus, 152-FZ, Article 16, 27 July 2006 (consultant.ru)
  6. GitHub, hhru/api, 2026 (github.com)

Found a mistake? Tell us

Publications

Contact us