Somewhere on an offshore wind platform, a technician hunches over a turbine gearbox at 5 AM. It is dark. The North Sea wind howls through open hatches and claws at his jacket. He needs to verify a bolt torque specification buried within 1000’s of pages of user manuals. There is no cell signal, no Wi-Fi, and no satellite link. Until recently, his only option was to guess, frustrated by hours of searching through emails and SharePoint, or lug a binder the size of a cinder block up 100 meters of ladder.
That changed when a small Amsterdam-based company called Twindo put a neural network on his phone. Twindo, led by co-founder and CEO Jules Shertser, has constructed the first commercial offline AI technician copilot that operates directly on a mobile device. The tool runs a custom Small Language Model on iOS, requiring only 4GB of RAM. It logged a 100% satisfaction rate and 93% accuracy during field testing.
Those numbers rival those of lower-end tier 1 large language models running on cloud servers costing millions. Google noticed. The tech giant invited the Twindo team to its Paris HQ and, for 12 weeks, collaborated on a special AI program designed to advance the energy space. Google has still not replicated a working copilot with similar capabilities on Android, despite the same RAM requirements. Twindo already has it live on iOS.
A Sensei In The Technician’s Pocket
The copilot does far more than surface keyword matches. A traditional search delivers raw text. Twindo’s model reasons through thousands of pages of enterprise documents and applies industry-specific context to produce precise, sourced answers. Technicians speak their questions aloud through voice-to-text. The response arrives in seconds, walking through the process and pinpointing the exact page where the data lives to verify accuracy.
“It sounds cheesy, but we see field technicians as giants. They work 12+ hour shifts, 7 days per week, on multi-week rotations, standing on these massive assets in the harshest environments on Earth with little to no support. Our technology stands on their shoulders, providing the support they need, which benefits the whole value chain,” Shertser said.
Shertser once climbed those towers himself. He spent over a decade as a hands-on wind energy technician and engineer before co-founding the company in 2022, months away from home, broken bones and all. That frontline experience gave Twindo an edge no Silicon Valley AI lab can buy. He knows what field workers actually need versus what software architects assume they want. The copilot was refined through rage click tracking and behavioral data, molded into something blue-collar workers willingly adopt, a rarity in heavy industry software.
Closing The Loop From Field To Office
The offline copilot represents one half of a dual-copilot system. A second, independently developed neural network serves management and office roles. Users query their operational data, trigger agentic workflows, and pull complex analytics within seconds. Twindo calls the concept “Vibe Inspections,” a term it coined to describe a new category where prompting replaces clicking.
The two copilots feed each other. When technicians capture richer data in the field, such as photos, videos, checklists, and inspection reports, the office copilot analyzes patterns across projects. Where are teams struggling? Which assets keep failing? How are timelines slipping? Over 40% of energy asset failures trace back to field human error, and Twindo’s ecosystem tackles that statistic head-on.
The platform already runs over 21,000 wind turbines worldwide, roughly 5% of the global fleet, including in China. Global clients trust the platform, and the company recently became the Dutch market leader in offshore wind field inspection software. Twindo achieved all of this with a team of 10, zero client churn, and a projected 2026 ARR above €1 million.
What Happens When The Signal Drops
The cloud-only AI paradigm carries a blind spot few talk about: the places where signals vanish. Transmission lines in remote valleys. Solar farms in arid plains. Offshore rigs surrounded by trashing saltwater. These are precisely the sites where critical decisions carry the highest stakes, and where Twindo’s offline-first philosophy lands hardest.
“The world’s appetite for AI compute is devouring energy faster than grids can supply it, and the workers building and maintaining those energy assets are stretched thinner than ever.” Said Shertser. “We help the companies doing that work move faster and cheaper.”
Twindo recently earned ISO 27001 certification, reinforcing its credibility with enterprise clients wary of entrusting AI-powered mission-critical operations to a startup. The company launches its asset-agnostic platform in Q2 2026. That opens the door beyond wind to solar, oil and gas, BESS, and transmission, a direct challenge to incumbents like SafetyCulture, ServiceNow, and Procore. Shertser and his team are raising a seed round to accelerate Android development and push GTM into new regions and energy verticals.
Whether Twindo becomes the default operating layer for the energy industry’s frontline depends on execution. Based on their capital-efficient achievements so far, this is entirely possible. After all, the company has proven something that OpenAI, Anthropic, and Google have yet to ship: intelligence that works when the internet does not. And for the technician standing on a giant at 5 AM in the dark, that might be the only thing that matters.
