Shibumi designs, builds, and ships machine-learning systems — tabular data science, computer vision, and AI-native products. We embed with your team and take you all the way to production.
What we do
Three areas of deep expertise, brought together in one team that can choose the right approach — and combine disciplines when the problem demands it.
Models that move the needle.
From business data to predictions and decisions.
Vision systems built for the real world.
Detection, segmentation, OCR, and tracking — on images, video, and live streams. Shipped to edge devices or the cloud.
Beyond the model: the system that makes AI work.
We build AI features and products around your domain, data and production constraints — from the harness and evaluation layer to model adaptation, deployment and optimization.
Our story
Shibumi (渋み) is a Japanese idea: beauty that is understated and effortless - complex underneath, simple on the surface. That's how we think good AI should feel.
We started in 2016, years before generative AI became mainstream, building machine-learning systems for real product and business problems. Since then, the tools have changed - from classical ML and computer vision to foundation models - but our role has remained the same: turn difficult problems into reliable systems that work in production.
Years and dozens of projects later, the mission hasn't changed, only the tools. From classical ML to computer vision to generative AI, we do one thing: turn hard problems into systems that ship. That's what we mean by Applied AI, delivered.
We prototype early and let real data, honest evaluation and working software guide the decisions.
We sit inside your team, learn your domain, and leave your engineers more capable than we found them.
A model in a notebook is a demo. We ship to real users, with evals, monitoring, and a handover.
Our vision
Applied AI shouldn't be locked inside big-tech labs. We're here to make world-class AI practical and production-ready for any company — whatever its size or sector. That's what Applied AI, delivered means to us.
World-class AI without a world-class research budget. We bring the frontier to teams of any size.
Models that ship and earn their keep — measured in business outcomes, not demos or slideware.
Systems you understand, own, and can rely on — with the evals and guardrails to prove it.
Selected clients
We’ve worked with startups, established companies and product teams across Data Science, Computer Vision and AI Systems Engineering.
Working with Shibumi has been a truly enjoyable experience — their expertise in machine learning is undeniable. They built us a driving simulation embedded with predictions, delivered exceptional quality, and responded to every need with real attention. We will undoubtedly engage them again.
Gidi is a real ML professional with a broad understanding of the field — genuinely good at finding the right solution to every problem, and very up to date with the latest models. I'd definitely choose Shibumi for another project.
We engaged Shibumi to bring computer vision into our security product. Their team built a tailored solution and guided us through data collection, development, testing, and integration. We really feel they're part of our team.
A highly experienced, knowledgeable, committed and professional team. They understood the project goals from the get-go and set a clear plan for execution — out-of-the-box thinkers who help you get there faster and better.
It's always a pleasure to work with Shibumi. They understand our requirements and quickly deliver high-quality results, using the latest developments in machine learning.
Gidi spotted what made our niche unique and ran a literature review that streamlined the whole effort. The response was fast and high-quality, and the solution works. Highly recommend.
How we work
Every engagement runs the same loop. It's deliberately lightweight — the goal is shipped software, not process for its own sake.
A short, sharp discovery. We pressure-test the problem, the data, and whether AI is even the right answer — before anyone writes code.
A working proof in weeks, not quarters. Real data, honest metrics, and a clear read on what's feasible and what it's worth.
We harden the winner into a real system: pipelines, evals, monitoring, and the unglamorous engineering that makes it last.
We work inside your team and leave it stronger — documentation, training, and ownership transferred so you're not dependent on us.
Let's talk
Tell us what you’re trying to build, improve or scale. We’ll help determine whether the bottleneck is the data, the model, the harness or the product around it.