A bot screened 1,518 hh.ru vacancies and sent 339 applications
Бот разобрал 1518 вакансий на hh.ru и отправил 339 откликов
B · Useful, limited A Python developer documents his own LLM job-application bot with full funnel data matched to hh.ru statuses; carefully reported, but one person, one CV and small samples per period.
Summary based on the full text·Original: about 6 minutes to read
In short
A Habr case (20 October 2026) by a Python developer with about two years of experience who looked for remote LLM and RAG roles in summer 2026. He built a bot that searched hh.ru, scored each vacancy with an LLM via Codex CLI, wrote a tailored cover letter and applied. Between 17 June and 17 August it screened 1,518 vacancies and sent 339 applications. He matched every application to its hh.ru outcome and to the bot version that sent it, so the post shows how each change moved conversion.
Key numbers
- 26 of 339applications sent by the bot that led to an invitation (7.7%)
- 4.6% to 23%invitation rate before and after adding seniority filters and a higher score threshold (152 and 26 applications)
- 2.9%invitation rate after widening the search and lowering the threshold for volume
- 18 of 31live recruiter chats in which an hh.ru robot recruiter or employer AI assistant ran a mini-interview
- 0 of 26invitations from applications that required an hh.ru test or questionnaire
As reported by the source.
Russian·Web page·opens habr.com
What it reports
- Funnel: 26 invitations (7.7%), 133 rejections (39.2%), 110 opened without reply (32.4%), 70 never opened (20.6%), across 258 companies.
- Rejections are fast: median 19 hours, a quarter within 1.5 hours.
- By bot version: first version 4.6%; seniority filter and threshold raised from 60 to 75, 23% (26 applications); stricter scoring and fact-only letters, 10.3%; letter opening with the candidate's own matching experience, 14.3%; wider search, lower threshold and 12 applications per run, 2.9%.
- The first prompt gave 100 out of 100 to 96 of 152 vacancies applied to in June, because it scored topic relevance and ignored seniority.
- After a scale was added, model scores predicted invitations only weakly: 8.0% for scores 60-79, 11.9% for 90-99.
- Employer-side automation: in 18 of 31 live chats an hh.ru robot recruiter or a company AI assistant ran a short screening interview.
- Applications with an hh.ru test: 26 sent, zero invitations by status.
- Sending time made no clear difference (9.1% morning versus 7.4% afternoon). The author found his job on another site, where replies came the same day.
How they got it
The bot logged every scored vacancy and sent letter in SQLite. Outcomes were scraped on 20 October from the JSON embedded in hh.ru's "Responses and invitations" page, including the archive, and joined to the bot log by vacancy id. Periods are split by commit dates of bot changes.
Limits
- One person, one CV, one role family; the author says samples per period prove nothing on their own.
- The August drop may reflect a quieter market.
- Invitations sent only through chat may be missed; timing of invitations could not be measured.
Why it matters for companies
Both sides of hiring are now automated: candidates run LLM bots that apply at scale, and employers answer with robot recruiters that pre-screen in chat. The case shows that precision of targeting, not volume, moved results, which is the same lesson for recruiting teams designing their own screening. It also shows that an LLM can judge topic fit but not whether a given employer will invite a given candidate, which limits fit scores as a screening signal.
Key passages in translation
Russian original
Самая полезная правка за все лето оказалась самой скучной: перестать откликаться на вакансии не своего уровня.
The most useful change all summer turned out to be the most boring: stop applying to vacancies above or below my level.
Russian original
Модель хорошо понимает, про ИИ ли вакансия, и гораздо хуже понимает, позовут ли туда меня.
The model judges well whether a vacancy is about AI, and much worse whether they will invite me.
Russian original
После отклика в чат часто приходил робот-рекрутер hh или ИИ-помощник компании и устраивал мини-собеседование
After an application, an hh.ru robot recruiter or a company's AI assistant often came into the chat and ran a mini-interview.
Summary based on the full text. Read the original for the full argument and data.Read the original on habr.com
- Tags
- Labour market