LinkedIn built an artificial intelligence (AI) infrastructure stack around agent workflows, moving beyond standard generated text capabilities to autonomous AI agents that can manage complex recruiting tasks.
This technology is currently being used to support the global rollout of LinkedIn Hiring Assistant. The Microsoft-owned company says the product already saves recruiters an average of four hours per role and reduces candidate profile reviews by 62%.
Prashanti Padmavan, vice president of engineering at LinkedIn Talent Solutions, told Computer Weekly in a recent interview that her company’s infrastructure stack is evolving to support the agent era, where AI not only summarizes text but executes multi-step workflows.
“We removed all the manual, labor-intensive parts of the recruitment process and used a combination of agent flows and a multimodal agent-based architecture,” says Padmavan.
Padmavan said LinkedIn Hiring Assistant is built on the LangGraph agent orchestration framework, allowing AI agents to do the heavy lifting of scouring LinkedIn’s database of 1 billion members to identify candidates for hiring teams.
The move comes as the recruitment market faces increasing friction. Approximately three in four recruiters in major markets in Asia Pacific say it has become significantly more difficult to find qualified candidates, according to new data released by LinkedIn.
“We are harnessing the power of LLM [large language models] “We process large amounts of data to come up with the right candidate for a given role,” Padmaban said. We also provide explanations and evidence as to why we believe these top candidates are the best fit for the role. ”
However, standard off-the-shelf LLMs are not effective in addressing the nuances of specialized company employment. Instead, LinkedIn uses an ensemble of fine-tuned models based on a vast dataset of skills, career changes, and professional relationships.
This allows AI agents to interpret natural language requests, such as a recruiter conversationally describing a role, without forcing them to create complex Boolean search strings.
“It requires a lot of domain-specific intelligence, and we’re using our own secret sauce of data and insights to make sure we can put it into context,” Padmaavan said, adding that the engineering team employs search augmentation generation, reinforcement learning, and other techniques to continually improve model performance.
For enterprise CIOs and HR leaders, integrating AI agents into existing HR management applications is necessary to help companies manage the entire employee lifecycle, from recruitment and onboarding to engagement and training.
Padmavan said LinkedIn’s recruiting capabilities can be integrated with popular HR platforms such as Workday and SuccessFactors as well as applicant tracking systems (ATS) to blend LinkedIn profile data with a company’s candidate records.
However, automation comes with the risk of AI-induced hiring bias. Padmavan emphasized that LinkedIn takes a human-involved approach, meaning that AI agents provide evidence and reasons for their selections, but recruiters make the final decision.
“We’re not letting hiring assistants make autonomous choices; their job is to do the hard work and communicate why they chose candidates with hard evidence,” she said.
She added that before agent AI products are deployed in production, they must pass a series of tests by a responsible AI team to check for security vulnerabilities such as gender and demographic bias and prompt injection attacks.
Early adopters of LinkedIn Hiring Assistant have benefited from the technology, including United Overseas Bank (UOB) and blockchain technology company OKX.
Speaking at the LinkedIn Talent Connect event in Singapore, Jay Chan, Executive Director and Head of Talent Acquisition at UOB, whose team is a charter customer of LinkedIn Hiring Assistant, used the tool to help identify specific candidates for eventual hire and justify return on investment (ROI) for business leaders.
Tracy Mao, director and head of human resources for Singapore and Malaysia at OKX, reported that the tool’s candidate matching capabilities outperformed manual outreach, saving recruiters about six to eight hours of time. However, Mao noted that the tool currently faces challenges in “combining with multilingual candidates.”
