Skip to main content
veles

// AI Agents · Recruiting

AI agents for recruiting that pre-screen candidates, not spam them.

Most recruiting AI is a glossy CV parser bolted onto a generic ATS. A custom AI agent owns the candidate conversation end to end: first-touch outreach, qualifying questions, calendar scheduling, status updates. Recruiters work the conversations that matter. The agent absorbs the rest.

The problem with most "AI recruiting tools"

Every ATS vendor has shipped an "AI feature" in the last two years. CV parsers, candidate-rank widgets, mass-outreach autoresponders. Most of them sit one layer above the actual recruitment workflow and produce more noise than signal: a 500-candidate ranked list nobody reads, an autoresponder that scares qualified candidates off, a parser that misses the one skill the role actually needs.

Meanwhile recruiters still spend the majority of their day on conversation work. Initial outreach to a sourced profile. Qualifying questions on a fresh application. Scheduling a screen across three time zones. Status updates to candidates waiting on a decision. The recruiter is the bottleneck, and the tools around them do not absorb the conversation load. They generate more of it.

The cost of slow recruiter response is well-documented in industry research. SHRM benchmarks put the US average cost-per-hire at $4,129 per non-executive role (SHRM Benchmarking Report), and a meaningful slice of that goes to the time-to-hire delta that response speed drives. Faster candidate conversations, lower cost-per-hire, less ghosting.

How we build it differently

We ship a vertical AI recruiting agent built around your specific roles, your ICP, your ATS, and the channels your candidates actually use. The agent reads every inbound message (application form replies, LinkedIn DMs, email, WhatsApp where applicable), classifies intent (application, follow-up question, scheduling request, status check), pulls candidate history from your ATS, and either drafts a response for recruiter approval or sends it straight through under rules you define.

Routine cases flow without recruiter intervention: scheduling first-screens on a published calendar, sending status updates after pipeline stage changes, answering questions about the role from your job description. Edge cases (unusual experience, salary negotiation, technical role-fit questions) escalate to the recruiter with a one-screen summary of the candidate, the conversation so far, and a suggested response.

Tool use over pure prompting is a hard rule. The agent does not "decide" who is qualified, it calls a scoring function we wire to your ICP definition. Same for calendar availability, ATS status writes, and email send. Auditable, deterministic, and impossible to hallucinate a candidate into a phantom pipeline.

Multilingual support is native. Frontier language models handle EU languages out of the box, including code-switching between English and the candidate's preferred language. We have shipped multilingual systems for cross-border recruitment teams.

What we ship for a recruitment client

  • Conversation reader: agent classifies intent on every candidate-side message (application, scheduling, status check, off-topic)
  • Candidate memory: every applicant recognized by application history, role context, and prior conversations
  • ICP scoring function: structured scoring against the role's real qualification criteria, returning a deterministic score the agent can quote
  • Calendar scheduler: first-screen booking on recruiter calendars with timezone handling and rescheduling support
  • ATS write-back: status updates, candidate notes, pipeline stage transitions sync into your existing ATS without manual data entry
  • Recruiter dashboard: every active conversation visible in real time, one-tap takeover, one-tap return to the agent
  • Multilingual conversation: native EU-language handling with code-switching support
  • Optional: pre-screen mock interview that runs before any human recruiter time gets booked

// Proof from production

Already running in production

Recruitment OS

People Up

We built exactly this shape for People Up, a recruitment firm placing talent across the EU and Canada. Their workstation handles candidate sourcing, real-time AI research, ICP scoring, mock-interview pre-screening, and email campaigns from one machine instead of five tabs. Recruiters open each candidate and see identified gaps, verified strengths, and suggested questions, all surfaced by the AI before the call. Live in production at peopleup.llc. The same architecture extends to any recruitment team whose pipeline has outgrown manual tracking.

Case study

// FAQ

Frequently asked

Does this work with our existing ATS (Greenhouse, Workday, Lever, etc.)?
Yes. The agent integrates via your ATS API or webhooks, not as a replacement. We have shipped against the major ATS platforms used in EU + North American markets. The agent reads candidate state from your ATS, writes status updates and notes back, and never owns the candidate record. If you switch ATS in two years, the agent moves with you.
What channels can the agent operate on for candidate communication?
Email is the most common entry point. LinkedIn DMs at sourcing scale (where allowed by platform policy), WhatsApp for markets where it is the dominant business channel (large parts of Europe, Asia, LATAM), and embedded chat on your careers page are all standard. The agent handles multi-channel conversations as one identity per candidate.
How does interview scheduling actually work?
The agent reads your recruiters' calendars (Google, Outlook, Calendly, or direct API), proposes the right slots given role, candidate timezone, and recruiter availability, and writes the confirmed slot back into both calendars. Reschedules, cancellations, and no-show recovery are handled inside the conversation. The recruiter does not touch the scheduling unless something genuinely unusual surfaces.
Can the agent pre-screen candidates before recruiter time gets booked?
Yes, and this is where the leverage really shows up. We build optional mock-interview pre-screening modules that run async with the candidate (5-15 minute focused conversation tailored to the role), surface signal on language fluency, technical depth, and red-flag answers, and feed a one-page summary to the recruiter before the human screen. Hours of recruiter time recovered per week.
Does it work for technical roles or only non-technical ones?
Both. Technical roles get domain-tuned conversation patterns (the agent can carry a competent conversation about generic technical concepts without trying to pretend it is a senior engineer) and tool-calling for skills-screening tasks. Non-technical roles use a different conversation pattern but the architecture is the same. The recruiter still makes the hire decision; the agent makes sure the conversation up to that decision is fast and structured.
How long until I see results?
A focused single-ATS recruiting agent ships in 4-6 weeks from kickoff to production, depending on integration complexity. Full scoping and milestone breakdown lives on our process page. First 2 weeks post-launch are close observation: we watch the agent handle real candidate flows, tune scoring criteria, and adjust escalation thresholds. By month two the agent owns the routine work end to end and the recruiter's day looks meaningfully different.

How we estimate the work

Pricing follows scope, not a fixed rate card. The full breakdown of how we scope, ship in milestones, and what each phase includes lives on the process page.

See our process →

Ready to talk about a recruiting project?

A discovery call is free and runs about 30 minutes. We map the problem, tell you if we are the right fit, and walk out with a first-milestone outline.

We use cookies to analyze site traffic and improve your experience. Privacy Policy