A full stop is a tiny thing to build a career around. It can end a sentence. It can sit inside an abbreviation. Give it a capital letter for company and a computer has another decision to make: has somebody begun a new thought, or merely written a name? Andrei Mikheev was working on these awkward little distinctions long before recruitment software began advertising its artificial intelligence. The punctuation had been causing trouble all along.
In 2002, the year he co-founded Daxtra, Mikheev published a paper called “Periods, Capitalized Words, etc.” Its title has the modest air of a drawer label. Its subject is less modest: how to turn the marks people put on a page into information a machine can use. A document, his research argued, carries clues in local context and repeated words. Read those clues together and an apparently trivial mark becomes intelligible.
That is a useful place to begin with Mikheev. His career connects the close reading of language with the practical business of hiring. The connection runs through research papers, a partnership with Steve Finch, a chance encounter with a recruiter and a company whose software would eventually handle CVs across dozens of languages. Through it all, the question keeps returning: what does this person actually mean?
The partnership before the company
Mikheev began his research career at the Russian Academy of Sciences. His doctoral work dealt with Russian morphology, syntactic parsing and acquiring knowledge from bodies of text. These are different ways into the same difficult territory: words change their forms, take their places in sentences and acquire meaning through use. A machine needs a method for following them there.
He joined the University of Edinburgh in 1993, focusing on machine learning applied to language technology. The setting matters because the later business grew out of research, rather than a recruitment firm deciding to commission a clever search box. Before there was a Daxtra sales proposition, there were questions about how language could be analysed at all.
Finch appears beside Mikheev in the research record well before the company’s founding. They co-authored a 1995 paper on a workbench for acquiring domain knowledge from natural language. In March 1997, their “A Workbench for Finding Structure in Texts” appeared at the Fifth Conference on Applied Natural Language Processing in Washington, DC. The future co-founders already had a shared subject and a habit of publishing together.
A workbench is a pleasingly unromantic image for AI. You bring material to it. You need tools. You expect to do some work. The titles place Mikheev and Finch in a tradition of making language processing usable, with all the adjustments that usability requires. Their later company would give that work a particular customer and an unusually untidy pile of documents.
What to do with a word you have never met
In another 1997 paper, Mikheev tackled unknown words. A language system may have a dictionary and still meet something outside it: a specialised term, an unusual form, a word its designers simply failed to anticipate. Refusing to proceed would make for a remarkably unhelpful reader.
His approach learned rules from a general-purpose lexicon and word frequencies in raw text. Beginnings and endings supplied evidence about possible parts of speech. The system could use regularities in known language to make sense of unfamiliar material, then pass those possibilities into further analysis. It was a measured kind of guessing, with rules evaluated rather than accepted on charm.
In 1999, Mikheev, Marc Moens and Claire Grover examined another assumption: that recognising names required extensive lists of them. Their experiments combined rules and statistical models, comparing gazetteers of different sizes. For the particular competition material and task they tested, relatively small lists of familiar names could suffice. The qualification belongs with the finding. Research earns its usefulness by specifying the conditions.
These papers help explain the intellectual route to recruitment technology. A CV introduces names, organisations, qualifications and job titles, often in combinations a system has never seen. The document cannot be made useful merely by recognising familiar strings. A reader has to work out how those strings relate to one another. Mikheev’s earlier subjects sit close to that everyday requirement.
A recruiter gives the research a destination
The commercial turn came through Sergei Makhmodov, whose background was in recruitment. Mikheev and Finch had developed software that could read and summarise written material. A chance meeting with Makhmodov opened a specific possibility: adapt it to parse CVs and résumés. Recruitment supplied a task for capabilities that could have travelled into several industries.
There is something instructive about that encounter. Researchers could understand the machinery of text; a recruiter could recognise the cost of processing it by hand. Together, those perspectives made a product imaginable. Daxtra was founded in 2002, and its Parser and Capture products developed with input from recruitment professionals. The technical problem acquired a working day, a user and a reason to be solved.
The company’s emphasis on employing people with recruitment experience continued that arrangement. A useful tool has to fit the decisions and routines around it. A technically elegant output can still leave somebody copying, correcting or searching again. Domain knowledge, in this setting, includes knowing what happens at the recruiter’s desk after the software has finished.
The reference is a different person
Mikheev has used a wonderfully mundane example to explain the difficulty: the references on a CV. A program that locates a name and address still has to determine whose details it has found. The candidate and the referee may share the same page. Giving the recruiter the wrong person’s contact record would be an efficient way to save no time whatsoever.
The same difficulty attends a word such as director. It can describe a corporate position, a film-maker or a software product. Document structure and nearby language help establish which meaning belongs. In recruitment, those distinctions affect who gets retrieved and what a system believes they have done. The consequences begin well before anybody reaches an interview.
By 2018, Mikheev was writing about the gap between a candidate naming an exact qualification, such as an MBA, and an employer expressing a broader educational requirement. Matching those descriptions requires a method for relating different language. He called the task a “scientific challenge”, placing methodology ahead of implementation.
His outlook also included Asian languages and the recruitment databases needed to support new tools. Chinese, Japanese and Korean could not be treated as decorative additions to an English-language system. The work required attention to their structures. The international business and the linguistic research were entangled: serving another market meant learning to read its documents properly.
Above the office, another kind of classroom
A Richmond visit in 2017 offers a glimpse of Mikheev away from papers and product descriptions. During a trip to Daxtra’s US office, he attended a rooftop fundraiser for CodeVA, the organisation working to bring computer science education to Virginia. He spoke with its founder, Chris Dovi, about computer science standards in education.
The photograph is agreeably ordinary: a brick parapet, a city roof, two people in conversation. It adds a different scale to a career often described through software. There are documents to be processed and businesses to be run; there are also discussions about how the next generation will encounter computing. Here, the researcher and executive is simply another participant at an education fundraiser.
A new job inside the company he built
In June 2023, Strattam Capital announced a majority growth investment alongside Mikheev and Finch. The stated purpose included meeting demand in North America. After more than two decades of company building, Daxtra was bringing in capital for another stage of expansion.
That October, Terry Baker succeeded Mikheev as CEO. Mikheev was to continue on the board and become Chief Innovation Architect. The announcement put some numbers around the business he had helped build: more than 2,500 clients had used Daxtra to evaluate over one billion résumés in 40 languages during the previous year. Those are historical figures, attached to that moment in the company’s development.
“I am looking forward to working closely with Terry as I focus on building innovative products.”
Andrei Mikheev, October 2023
The sentence makes a fairly clear statement of intent. His next assignment centred on products. A founder’s change of title can invite elaborate interpretations; Mikheev’s own words point towards work at the drawing board. The research questions that helped start the company still had commercial life in them.
Still reading between the keywords
By August 2026, his title on a co-authored Daxtra article was Chief Architect and Innovation Officer. With marketing executive Paige La Fever, he returned to a distinction familiar from his earlier work: retrieving CVs with similar text and evaluating candidates against a role are separate operations.
The piece argues for examining seniority, skills, certifications and industry experience as individual dimensions, with explanations accompanying rankings. Read against his earlier career, it is a continuation of the same preoccupation. Useful information lies in relationships and context. A page can contain the expected vocabulary while describing a person whose experience answers a different question.
Mikheev’s story stretches from the form of a Russian word to the structure of a hiring decision. Its appeal is in the continuity. New methods arrive, companies expand and executive titles change. Meanwhile, somebody still has to teach the computer why the name at the bottom of the CV belongs to the reference. The full stop may be tiny. There is plenty of work on either side of it.