Vasilii Shelkov Took the Long Way Down

In late-1980s Moscow, the lights seemed low and the future had not yet decided what it wanted to be. Vasilii Shelkov remembers the city as literally dark, a place where finding a bar could become an expedition. Yet in Dubna, the science town where his family lived, computers were already glowing. Both his parents were scientists. Their work gave their son access to mainframes, and he was sending email before university, when most people had never seen an inbox.

He watched his father and his colleagues build particle detectors, devise experiments and compete for the next discovery. The work looked like fun. Physics was less an inheritance than a household climate, and Shelkov breathed it in. In 1987 he entered the Moscow Institute of Physics and Technology to study high-energy physics and applied mathematics. The Soviet order was loosening around him. A compulsory class on Communist Party history disappeared from the curriculum and, in the institutional scramble that followed, students received an overview of world religions instead. Shelkov tells the story with a laugh. A system had removed one certainty and filled the timetable with many.

The opening also brought Pizza Hut. He took his future wife there on their first date, joining the queues that formed for a first taste of a Western chain. It is a modest detail, but a revealing one: large political changes tend to arrive in private life as a menu, a queue, a possibility that was not there last year.

Texas, before the oil

In 1993, Shelkov moved to Dallas for a physics doctorate at Southern Methodist University. Texas was then betting on the Superconducting Super Collider, a machine intended to exceed anything operating at CERN. Two miles of tunnel were built before Washington cancelled the project. For a young physicist who had crossed an ocean to join the excitement, the anticlimax might have felt cosmic. Instead, the work shifted. Shelkov went to Cornell, where the CLEO experiment had a functioning collider and unanswered questions.

His dissertation examined tau-lepton decays, the brief traces left by an elementary particle that exists for a fraction of a second. He found an unusual decay pattern that had not previously been observed. It did not overturn the Standard Model, he later said, but it did help open the next door. In 1997 he earned his Ph.D. and joined the BaBar group at Lawrence Berkeley National Laboratory, with much of the work based at SLAC near Stanford.

The Berkeley years supplied the technical grammar of his later life: neural networks, data analysis and C++. Collider experiments generate torrents of indirect evidence. The software has to sift signal from noise, coordinate enormous collaborations and remain honest about uncertainty. Reservoir simulation asks different questions, but it shares the same temper. The object of interest is hidden. Measurements are incomplete. The physics matters. So does the speed of the calculation.

“It got me into the very things that I’m still doing today.”Vasilii Shelkov on learning neural networks, data analysis and C++

Berkeley also offered a close view of startup culture. Shelkov called the place a “giant soup” of government, private and military money, hardware and restless people. After six years and two postdoctoral appointments, he was ready to leave the collider cycle. A move to Brookhaven National Laboratory on Long Island followed, but the work felt familiar in the wrong way. The community was smaller than it had appeared from inside it.

Nine months that changed the scale

Friends pointed Shelkov toward a software job at Yukos in Moscow in 2004. There, an adjacent team was building an early simulator for monitoring water floods in oil reservoirs. He had only nine months in the industry when Yukos collapsed, but nine months were enough. He had met Kirill Bogachev and found a problem with the combination he knew well: dense physics, awkward data and an appetite for computation.

The displaced colleagues went to interviews at established service companies. The courtship was not warm. Shelkov felt they did not much like the candidates, and the candidates returned the sentiment. So in 2005, he and Bogachev co-founded Rock Flow Dynamics. The aim was to create reservoir-simulation software that felt modern in performance and in use. Bogachev became chief technology officer; Shelkov became chief executive. Their product became tNavigator.

Scepticism was plentiful. The established providers were large, installed and familiar. A physicist with less than a year of petroleum experience was an easy target for the question “Who are you to compete?” Shelkov’s answer was internal before it was commercial: a founder first has to persuade himself. Otherwise, he said, the road becomes hell.

Early sales were thin. Intel invited the young company to trade shows because its software ran on Intel hardware, then invested $2 million in 2010. The money mattered, but so did the endorsement. RFD had crossed what Shelkov called the bridge of initial negativity. It was still alive. Whenever cash came in, he says, the company hired more developers. His rule is deliciously unfashionable: do not hire salespeople when the product does not sell; perhaps the product needs improving.

The eight-core carry-on

One early episode compresses Shelkov’s operating style into a single week. He wanted Apple’s first eight-core Mac Pro, but it was unavailable in Europe and Russia. On Tuesday he flew to New York. On Thursday he queued at the Fifth Avenue store. By Monday, customers in Moscow were watching their models run on it. Procurement had become performance theatre, with a useful result at the end.

In 2010 he moved back to the United States and opened RFD’s Houston office. Within eight months, the team made its first sale to an exploration company in Dallas. The company then expanded across countries because the niche demanded it. Reservoir software has long sales cycles and a limited pool of customers. Even the United States or the Middle East alone was too small. When money was scarce, clients in another market kept the lights on. Internationalism was not a decorative value. It was working capital.

“I’m used to checking a new place for two essential things straight away: Is there a rack room for our computing cluster, and is the electricity supply sufficient?”Vasilii Shelkov

RFD later bought Intel out. The investor eventually needed a return; Shelkov and his colleagues did not want the company sold. They negotiated their independence and kept building. By 2023, company material described a global team of more than 300. In 2025, RFD marked its twentieth anniversary. The same year, it gave SMU access to its software for teaching and research in energy, geothermal, mining and geotechnical engineering, returning an industrial tool to the university where Shelkov had learned to make invisible systems measurable.

A software factory with a pulse

As tNavigator widened from reservoir simulation into geological, well, fracture, geomechanical and surface-network workflows, the codebase grew beyond 10 million lines. At that size, a quick fix can create a distant failure. RFD calls its continuous comparison of new code against established results regression testing. Shelkov calls it the heartbeat.

A new data-processing centre in Belgrade changed that rhythm. Tests that could take as long as three days came down to roughly eleven hours. The improvement sounds like infrastructure, but it changes the social life of engineering. A developer can learn sooner whether a fix worked. Clients wait less. Ambition becomes less expensive because failed ideas can be discarded before everyone forgets why they were tried.

Shelkov is wary of software that stops at attractive pictures. An algorithm, he says, must produce a result that makes sense. That insistence links the tau lepton to the reservoir. Both are invisible worlds reconstructed from evidence. Both tempt the modeller to admire the image and forget the test.

His horizon now extends downstream. He talks about bringing pipelines and refinery simulation into one system so production and economics can be modelled together. Geothermal, carbon storage and mining have widened the set of underground questions. Machine learning has returned too, no longer an obscure technique encountered at a California collider but a common promise in industrial software. Shelkov’s preference remains practical: put physics and computation behind the promise.

Carry on

Shelkov once compared running a company to a long-term relationship: give and take, but always carry on. It is a homely metaphor for a business built around supercomputers, yet it suits him. The career has been full of aborted machines, collapsed employers, sceptical buyers and slow markets. None produced a clean break. Each supplied a piece of the next thing.

There is wit in his public manner, sometimes with the brackets of an old-school smile attached. There is also impatience with ceremony. On LinkedIn, he has joked about skipping expensive system integrators and letting skilled internal teams assemble commodity hardware with open-source tools. The joke works because it is also an operating thesis: keep competence close, spend on the machine, learn how it works.

He arrived in Texas for a collider that was never completed. More than three decades later, he is in Houston, leading a company that models what lies miles beneath Texas and far beyond it. The grand machine vanished; the computational habit survived. Careers often look coherent only after someone draws the line. Shelkov’s line goes Moscow, Dallas, Berkeley, Long Island, Moscow again, Houston, and then outward. It is long, occasionally absurd, and pointed steadily down.