Recruiting · Production

LinkedIn Hiring Assistant: Multi-Agent Architecture for AI-Powered Recruiting at Scale

The problem

Professional recruiters spent most of their time manually reviewing hundreds of candidates, and the existing AI-assisted search (Recruiter 2024) still lost semantic information when translating natural language into constrained structured search filters, leaving the core time-consuming task unsolved.

First attempt

A single LLM block architecture for natural language search created quality scaling bottlenecks because fixing one prompt example could degrade unrelated outputs through LLM non-determinism, and translating open-ended natural language into constrained structured filter formats systematically dropped semantic information.

Workflow diagram · grounded in source
1
Recruiter submits hiring intent
trigger
“you're putting in your hiring intent. You could do things like attach a job posting, attach some notes you've made about the candidate you want to hire. Maybe you can even attach the resume of an employee who's leaving and who you need t…”
2
Supervisor agent routes request
routing
“The supervisor is going to determine, what is that intent about? What is the user trying to do here? Then based on that, it's going to send the request, possibly with some other information to another agent.”
3
Intake agent generates qualifications
ai_action
“the intake agent will take over. It might use some skills to look up user preferences, maybe things that the recruiter typically does, or other things associated with this hiring project. Then, based on that, it can generate a set of qua…”
4
Sourcing agent runs parallel searches
ai_action
“our sourcing agent, which is going to read those same qualifications we just generated. Then it's going to use other skills like the ones we talked about to generate search queries, and then feed those through our Recruiter Search index”
5
Candidate evaluation with citations
ai_action
“for each candidate, taking the qualifications, the profile and resume of the candidate, and then producing an evaluation that says, for each of the qualifications, whether the LLM was able to find evidence for that. If so, it gives you c…”
6
Human-in-the-loop escalation
human_review
“We have an explicit concept of human in the loop, where if the agent cannot do something, it escalates to a human. After that, the human responds. The human is actively in the loop.”
7
Experiential memory update
feedback_loop
“we also have this cool thing called experiential memory, where the agent learns from its experience of interaction with the human and it stores things”
Reported outcome

LinkedIn built a multi-agent Hiring Assistant that evaluates candidates against natural-language qualifications with per-qualification citation evidence, runs multiple parallel searches to explore the candidate space, escalates to humans when needed, and operates on a shared agent platform that also enabled a parallel SMB variant to be built from the same components.

Reported metrics
Parallel search query executionrun multiple different queries in parallel, explored the search space in different ways
Candidate evaluation transparencycitations that show you where that supporting evidence comes from on the profile or the resume
Reported stack
LangChainLangGraphGPT-4oAzure OpenAIEONLinkedIn Recruiter
Source
https://www.infoq.com/presentations/LinkedIn-agent-hiring-assistant/
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

LinkedIn built a multi-agent Hiring Assistant that evaluates candidates against natural-language qualifications with per-qualification citation evidence, runs multiple parallel searches to explore the candidate space,…

What tools did this team use?

LangChain, LangGraph, GPT-4o, Azure OpenAI, EON, LinkedIn Recruiter.

What results were reported?

Parallel search query execution: run multiple different queries in parallel, explored the search space in different ways; Candidate evaluation transparency: citations that show you where that supporting evidence comes from on the profile or the resume (source-reported, not independently verified).

What failed first in this deployment?

A single LLM block architecture for natural language search created quality scaling bottlenecks because fixing one prompt example could degrade unrelated outputs through LLM non-determinism, and translating open-ended…

How is this recruiting AI workflow structured?

Recruiter submits hiring intent → Supervisor agent routes request → Intake agent generates qualifications → Sourcing agent runs parallel searches → Candidate evaluation with citations → Human-in-the-loop escalation → Experiential memory update.

Related recruiting cases