Eylon Etshtein and the problem that followed him home

A company can teach cameras to pick a face out of a crowd and still struggle to find its next employee. Eylon Etshtein encountered that awkward little joke from the inside. At AnyVision, the computer vision business he co-founded, he became closely involved in evaluating candidates. The attention was personal. The arithmetic was unforgiving. A growing organization needed more people; its chief executive still had only one working day.

That hiring problem supplied the idea for Perfect, his AI recruiting company. It is an unusually revealing origin for a founder associated with machines that recognize people. The next challenge was to help a business decide whom it should meet. There is a considerable distance between recognizing a face and understanding a career. Etshtein’s working life now runs across both.

By February 2025, Perfect had announced $23 million in total seed funding. The money gave the new chapter a conspicuous number. The more interesting detail was the old irritation behind it: a chief executive’s useful habit had become a limit on what his company could do.

The education before the recognition

Before the computer vision company, there were websites and digital businesses. Etshtein’s early work included Puzzle Interactive and Emka Digital. The tools and customers were different from those he would encounter later, but the work already combined programming with the business of getting something used.

His account of those years includes soccer, organizing parties, coding and design. He attended the Technion, Israel’s Institute of Technology, while finding the things he could make more absorbing than parts of the formal curriculum. An early agency benefited from the period when businesses could build applications within Facebook. A merger brought experience inside a larger organization.

Military service also figures in his account as a period of structure and teamwork. These details give the early career some texture without requiring the usual childhood prophecy. Website development and party organizing are perfectly respectable places to begin. Both offer brisk feedback when the thing you have arranged fails to attract anyone.

A camera was only the beginning

AnyVision was founded in 2015, with Etshtein as its founding CEO. Its work centered on recognizing faces, bodies and objects through camera footage. The commercial attraction was practical: software could add capabilities to cameras that organizations already had. Customers would not necessarily have to replace their entire equipment inventory before they could try the product.

In July 2018, the company announced a $28 million Series A led by Bosch. Contemporary accounts put the workforce at 130, spread across Israel, the United States and Britain. It was a substantial operating business to assemble around research that could otherwise have remained an impressive demonstration.

Etshtein talked publicly about access to offices and stadiums. A key had become a card; facial recognition might become the next credential. At a security industry event, he described spectators wanting to enter a stadium without stopping to scan a barcode. For the person at the turnstile, the sales pitch was the disappearance of a small, familiar delay.

His responsibilities extended well beyond demonstrating a match. There were investors, partners, sales efforts and employees to recruit. Building a company around recognition meant persuading organizations to buy it and people to work on it. The camera could do its part while the founder’s calendar filled up anyway.

The questions beyond the camera

Facial recognition also made AnyVision a participant in arguments that could not be settled by a product demonstration. Reports about its use in the occupied West Bank brought scrutiny of the company and of Microsoft’s investment. The dispute concerned surveillance, rights and the purposes for which a recognition system could be used.

Microsoft commissioned an audit led by former US attorney general Eric Holder and his team at Covington & Burling. In March 2020, the companies said the available evidence did not substantiate the alleged West Bank mass surveillance program or a breach of Microsoft’s investment pledge. Microsoft nevertheless decided to divest and end minority investments in businesses selling facial recognition technology.

The explanation focused on oversight: a minority investor had limited control over how a sensitive technology was used. The findings and the divestment belong together in the account. Neither cancels the other. For Etshtein’s company, commercial growth had arrived with questions about who could inspect, direct and constrain what its systems did.

Later that year came a change in leadership. Avi Golan’s appointment as chief executive took effect on November 1, 2020, with Etshtein moving to president. The founding CEO publicly supported bringing experienced leadership into the role. AnyVision changed its name to Oosto in October 2021. Metropolis acquired it in January 2025, years after that handoff.

The search inside his own company

Etshtein’s next operating chapter took up a problem closer to his desk. He had seen how much recruiting depended on direct human attention, including his own. The idea for Perfect grew from that experience. The product would help teams find potential employees and manage the work around reaching them.

Ronen Daniel, a childhood friend who had worked with him at AnyVision, became a co-founder. Ofer Razon is also listed among Perfect’s founders. Those connections matter because the recruiting idea emerged from an existing working world, with people who had already encountered its practical demands.

The business announced $13 million in seed funding in 2022. Its February 2025 announcement brought total seed financing to $23 million, with Hanaco Ventures leading the latest financing and participation from Joule Ventures and former Samsung president Young Sohn. These are financing milestones for the company. They say something about the resources behind the experiment, rather less about whether any individual candidate will accept an interview.

Daniel’s departure post in May 2025 recalled four co-founders working in a backroom office and a team that grew beyond 30 people. The accompanying gathering is a useful interruption to the language of automation. A company selling recruiting software has its own colleagues, departures and shared tables. Etshtein replied with an affectionate farewell.

Before the chatbot arrived

“When we started Perfect, ChatGPT was not out,” Etshtein said in 2025. That timing complicated the easy assumption that every new AI application began by placing a chat window over an existing model. Perfect had spent roughly three years developing its platform in stealth, he said.

His stated ambition was to understand a candidate’s career across past experience, present position and possible next steps. A résumé is a record of where someone has been. Recruiting also asks whether they might move somewhere else. The distinction is easy to describe and considerably harder to turn into a useful list of names.

Etshtein has framed the intended benefit in human terms: “Our goal is not to replace humans but to make them superhuman.” Set aside the cape and the proposition is familiar. Recruiters spend time searching, organizing information and arranging contact. Software that takes on some of that work leaves them more room for conversations.

By February 2026, he was announcing inbound screening alongside outbound sourcing: handling applications arriving at a company as well as looking for people elsewhere. The product’s scope was spreading across more of the recruiter’s day.

A good match needs an explanation

In May 2026, a technical account of GoPerfect’s work with Qdrant described a persistent difficulty: people can look similar in a search system while having substantially different experience. A product manager and a product marketer make a neat example. Close words do not guarantee interchangeable jobs.

The team described combining keyword and semantic search, representing different aspects of each candidate separately, and using a language model to coordinate searches and assemble results. The point was to return candidates that survived practical scrutiny. A list that merely looks plausible hands the work back to the recruiter.

Etshtein had been describing a related problem in his own product writing. Recruiters initially liked being able to ask for people in ordinary language. Then they wanted to know what the system had assumed. Some began writing increasingly detailed prompts to reconstruct the hidden filters. The apparent convenience had acquired a detective’s workload.

“The moment an AI acts on your behalf, trust becomes a design problem.”Eylon Etshtein, April 2026

Perfect’s response was to expose search parameters in an editable panel. Users could see requirements arranged by priority and change the choices the AI had made. The feature gave a recruiter a way to correct the search without having to guess at its internal interpretation.

In April 2026, Etshtein also announced GoPerfect Labs, an effort to publish recruiting insights drawn from the platform’s own activity. He described plans to share findings about sourcing, outreach and screening. The proposal takes the company beyond returning names: it would also explain patterns in how recruiting works. The announcement was a publishing plan, with the first finding promised for the following week. It offered another expression of the same interest in looking inside the process, this time through aggregated behavior rather than the settings for a particular search.

That is where Etshtein’s story currently becomes most interesting. He began this chapter with an excess of work requiring his personal attention. Now he is dealing with the attention people still need to give a machine’s decisions. The recruiter wants relief from the search, and a view of how it was done. After all the recognition systems and financing announcements, the next useful product detail is pleasingly ordinary: let the person open the panel and change their mind.