Aurimas
Griciūnas

The Engineer Who Reads the Production Logs

When Aurimas Griciūnas says something about AI, he has logs to back it up. Not demos. Not benchmarks. Logs from systems he shipped, monitored, and fixed at 2am. That distinction - between someone who talks about AI and someone who runs it - is the whole premise of SwirlAI.

He grew up in Lithuania and studied Financial and Actuarial Mathematics at Vilnius University, which is a pretty specific way to enter the ML world. But the training held: probability, risk, the math of uncertainty. That's the lens he brings to every architecture decision he's made since.

His early career read like a tour of production ML before production ML was glamorous. Danske Bank, where he built affordability models for a financial institution with real regulatory skin in the game. Transactie Monitoring Nederland, where he engineered systems specifically designed to detect financial crime. Not POCs. Production systems that had to work every time, because the alternative was letting criminals through.

By the time he joined neptune.ai as a Senior MLOps Engineer, the ML community was starting to wake up to the gap between notebook experiments and real-world deployment. Aurimas had been living in that gap for years. He rose to Chief Product Officer - the person responsible for deciding what neptune.ai built and why. Then OpenAI acquired neptune.ai. The product he helped shape ended up in the most famous AI lab on earth.

The Newsletter That Waited

He launched the SwirlAI Newsletter in October 2022 - before the ChatGPT wave hit, before "AI engineer" was a standard job title, before every LinkedIn post ended with "thoughts?" The newsletter was about doing the work, not reacting to the hype.

Then neptune.ai got intense. He took a hiatus. The newsletter went quiet for just over a year. When he came back in October 2024, 34,000 subscribers were still there. They had waited. That is either loyalty or stubbornness, and in this case both are compliments.

Today the newsletter covers context engineering, evaluation-driven development, agentic systems, and LLMOps - the plumbing that makes AI systems reliable instead of just impressive. His readership has since grown to 58,000+ Substack followers, making him one of the most-followed technical voices in the ML engineering space.

His philosophy is deliberately unglamorous: enterprises will take another five years to fully utilize the technology that existed eighteen months ago. That's not pessimism - it's a reading of organizational physics. The hype moves faster than the infrastructure. Aurimas builds for the infrastructure.

Building the Thing That Actually Ships

The gap between AI demo and AI product is one of the most expensive gaps in software. Aurimas has spent a decade living there. His time at Tesonet, neptune.ai, TMNL, and Danske Bank wasn't about researching AI - it was about deploying it into environments where failure has consequences. Regulatory consequences, financial consequences, security consequences.

That background made him a natural fit for neptune.ai's CPO role. The product was built for practitioners - the engineers running hundreds of experiments and needing to track, compare, and reproduce results. As CPO, Aurimas had to understand both the tooling needs of ML teams and the organizational dynamics that determined whether those tools got adopted. He navigated that until OpenAI came along.

The acquisition of neptune.ai by OpenAI is a quietly remarkable footnote in Griciūnas's career. He didn't build the company - but he built significant parts of what made it worth acquiring. That's the operator's quiet contribution: not always the headline, but always in the code.

SwirlAI today operates as both education company and consulting firm. He runs an 8-week cohort-based bootcamp on Maven that takes engineers through RAG architectures, agentic systems, multi-agent workflows, LLMOps, and cloud deployment - with a real capstone project. Not a Udemy course. An actual curriculum with live cohorts.

What He Actually Thinks

Aurimas's views on AI are calibrated by a mathematician who's watched enterprise adoption cycles in person. When he says enterprises will take another five years to catch up to technology from 18 months ago, he's not dismissing AI - he's describing organizational friction. Integration cycles. Compliance reviews. The speed of procurement vs. the speed of research.

On team composition, he's precise: the most valuable AI specialist is the one who can own the full stack - from product intuition to POC to production to post-launch feedback loops. Not the narrowest specialist. The widest one who still goes deep enough to ship.

His newsletter topics tell you what he thinks is real vs. noise. Context engineering made the list early. Evaluation-driven development for agentic systems made it in 2025. These aren't trend-chasing - they're the problems that production teams are actually hitting.

He's also a visible advocate for the Baltic tech scene - based in Vilnius, member of the AI Association of Lithuania, and a consistent presence at Lithuanian and Central/Eastern European tech events. At Infoshare 2025 - the biggest tech conference in CEE - he spoke on the DEV TRENDS stage about challenges and opportunities in agentic systems. At the Vilnius AI Summit in April 2026, he addressed how evaluation creates improvement loops in agentic AI.

The Long Game

What He's Actually Building

SwirlAI today is two things running in parallel. There's the education arm - the newsletter, the Maven bootcamp, the YouTube channel, the GitHub repos - all aimed at training a generation of engineers who can build AI systems that don't fall apart in production. Then there's the consulting arm, working directly with organizations trying to move from AI pilot to AI product.

Both share the same thesis: the limiting factor in AI adoption isn't the models. It's the people who know how to deploy them responsibly and the organizational infrastructure to support that deployment. Aurimas is building on both fronts.

The aspiration he's operating toward is a specific one: train engineers who can build production-ready AI, and help organizations develop AI strategy that survives contact with the enterprise reality. He's seen too many demos. He wants to see more pipelines.

The Lithuanian Angle

It's worth noting the geography. Vilnius is not a default node on the AI industry network. Aurimas Griciūnas built his following, shipped his products, got his CPO role, and found his acquisition from Lithuania. He's a member of the AI Association of Lithuania and a recurring presence at regional tech events - not because he's stuck there, but because he's planted there.

The CEE tech scene is growing and Griciūnas is growing with it - as a participant and, increasingly, as a flag-bearer. When he speaks at Infoshare or the Vilnius AI Summit, it's not just a speaking slot. It's evidence that serious AI engineering can come from anywhere, including actuarial mathematics programs in Lithuania.

His story is the kind that doesn't fit the standard Silicon Valley arc. No dropout mythology. No Y Combinator. Just a mathematician who got good at running ML in production, wrote about it honestly, and found a surprisingly large number of people who were hungry for exactly that.