All About AI: Begüm Dogru Öztekin of Refabric

We’re observing a big disconnect (in both our own fashion-only data and in wider benchmarks from other sources) between the increasing capability and reliability of AI, in both a general sense and in discrete applications, and the lack of trust that end users place in its outputs. What does trust in your specific applications of AI look like, and how do you achieve and measure it?

For us, trust is built on consistency. Every image is preprocessed, every prompt is enriched, and every request is routed through a pipeline built specifically for fashion use cases. Every ingested image is analyzed down to its finest fashion-specific attributes, construction details, fabric behavior, trim, and stitch. This is not a generic vision model with a fashion layer on top; the pipeline is built for fashion from the ground up.

But we measure trust the way our customers do: in practical results. Fewer sampling rounds, fewer corrections, production-ready outcomes. We track exactly those metrics with our enterprise clients. When the output becomes reliable enough to act on without rework, trust follows naturally.

Most AI tools that touch design and / or development start with one or more off-the-shelf, industry-agnostic models, and they primarily begin with quick-win visualisation and sharing as their targets. Refabric started somewhere different, in the manufacturing and hands-on fashion community in Istanbul, and your roots are there rather than in machine learning.

Has that supply-side origin influenced what you’ve subsequently built, and the capabilities you’ve chosen to prioritise? Because it’s fair to say that visual-first AI tools are going to encounter a potentially larger gap to overcome between how something looks and what it takes to make it, than companies going the opposite direction.

Absolutely, and in a very direct way. Coming from a textile background, with hands-on experience in sourcing and production and working closely with large manufacturing companies, gave us something you cannot get from anywhere else: direct access to real operational problems. We were in the room when brands and factories were struggling with the handoff between design and production, and that is where the idea for Refabric actually came from.

But I’d add that the origin isn’t just mine; it’s reflected in how the founding team is built, and that combination is a real strength. I’m a textile engineer myself, with hands-on experience in sourcing and production, right where the industry’s real operational pain lives. My co-founder Seda Domaniç spent ten years as Editor-in-Chief of Vogue Türkiye, so she brings a decade of editorial judgement and creative authority from the very top of the industry. And my co-founder Safiye Turgut, our CTO, is a data scientist and machine-learning engineer who turns all of that domain knowledge into real technology. So as founders, we already cover the full arc the product has to span: engineering and the production floor, creative direction, and the machine learning underneath it. And that same blend runs through the wider team. We’ve deliberately built a group that pairs deep fashion and manufacturing expertise with serious technical talent, so the people building the product genuinely understand the industry they’re building for. That is rare, and it is hard to compete with.

That mix is what allowed us to evolve the product from its initial form into what it is today. Analyzing the industry’s needs directly was fundamental to understanding where the actual gaps were, not where people assumed the gaps were, but where the pain was felt in every collection. And we keep doing it, because staying that close to the industry is how we make sure we are building for its real need for innovation rather than for the version of the problem that looks good in a demo.

When we last interviewed you, for the AI Report 2024, you told us Refabric’s moat was its fashion knowledge, and how that translated into product at a time when the labs behind the biggest general purpose models had done very little to package them up with skills and connectors for specific industries.

That’s changed in 2026. Software development is the obvious reference case, but companies like Anthropic and OpenAI are now much more focused on making those big cloud models work in narrow, and even regulated, industries like banking and legal. It feels like only a matter of time before that happens for CPG and fashion as well.

Do you think it still counts for a lot today to be a fashion-specific technology provider?

It matters more than it did in 2024, but the moat has moved, and it’s worth being precise about where.

A foundation-model lab can absolutely build a strong general image model and wrap it in fashion-aware prompting. What they cannot easily acquire is the thing underneath our product: proprietary, structured data on how garments are actually constructed, how fabrics behave, and what they cost to make, gathered from years of working inside the production side of the industry. We have analyzed garments down to trim, stitch, seam, and grade. We hold structured data on pattern blocks and grading rules, how a base size is engineered and how it scales across a size run, which is exactly the layer that lets a design become a manufacturable spec rather than just an image. And we connect that to real fabric pricing and to country-level manufacturer cost analysis: what a given construction costs to produce in different sourcing geographies, factoring in fabric, labour, and the minimums and lead times that actually decide where a style gets made. That data, and the relationships that generate it, are not something a horizontal lab has a reason or a route to build.

So the features people see, one-click Brand DNA application, multi-pose photoshoot with face consistency and fabric preservation, image-to-tech-pack generation, are not the moat itself. They are the visible output of that underlying fashion data layer. A generalist tool can imitate a feature; it cannot imitate the data and the production knowledge that make the feature reliable at a brand’s quality bar. That gap is what fashion professionals are actually paying for: precision, consistency, and brand control that general tools do not offer.

What has being part of the LVMH La Maison des Startups revealed to you about how luxury, in particular, is thinking about AI for product design and creation? Because we see high-end companies pushing the envelope in advertising and eCommerce, but it’s trickier to imagine where that segment stands on letting it touch creativity at the product level. What does luxury see in AI that maybe the mass market doesn’t? And is there anything that cuts the other way?

What I’d say first is that luxury sets a higher bar, and that pushes us to build a better product. The standard a Maison expects, on brand fidelity, on craft, and just as critically on confidentiality and data security, is the standard that makes the technology genuinely useful and trustworthy for everyone else too. Working to that level is not a constraint for us; it is a design principle.  Luxury also has very specific dynamics. The creative component is far greater, and in the kind of Maisons you find within a group like LVMH, product development runs through a chain of people with deep creative interdependencies. Introducing AI into that is a different conversation from automating routine tasks. At the product level, there is simply more identity and craft embedded in the workflow, so it is harder, and rightly so, to make space for AI there.

Over our two years in La Maison des Startups, the attitude has shifted noticeably. The first year was more exploratory, curiosity balanced with a healthy caution. By the second year, the conversations had become far more concrete and outcome-focused. That move, from understanding what the technology is to working through where it can genuinely add value, is the most meaningful signal I can point to, and it is a shift I’d attribute to the whole ecosystem maturing rather than to any single brand.

Your promise is to power “the entire fashion workflow,” covering everything from patterns to final PDP images. Which stages of that journey are the ones where you see the greatest uptake, and which do you consider to be the least mature?

Perhaps more importantly, how do you focus your roadmap when you’re aiming for full-spectrum coverage?

Our strongest area is design, where our capabilities are genuinely unique in the market. Tech pack generation is the next, and it is where we are arguably furthest ahead of the field. Today we deliver tech pack documentation complete with measurement charts, 3D mock-ups, costing, and pattern blocks, turning an AI-generated design directly into factory-ready technical documentation rather than just a picture of a garment. Converting ideation into a manufacturable specification at that level of detail is something most players have not cracked, and we are continuing to invest heavily in our pattern blocks to make that output sharper and more production-accurate over time.

We do face direct competition in some of the more visual, downstream stages, and I won’t pretend otherwise. But our differentiation there is never image quality alone. It is precise control, consistency across a full collection, brand-standard compliance, and fidelity to the actual garment at scale. That combination is far harder to replicate than a single impressive image, because it depends on the same underlying production data, not on a better-looking render.

On the roadmap, our principle is simple: our entire focus is on taking ideation and turning it into production-ready output. So we prioritize the stages where the cost of getting it wrong is highest and where we hold a data advantage others cannot match. That is why design and tech pack lead. We extend into adjacent stages only when we can bring that same production-grade reliability, rather than spreading thin to claim coverage.

Last year, we asked technology executives whether they predicted AI would become more obvious, as a primary interface paradigm, or whether it would disappear into the background in the way that vital but invisible platforms like AWS and Azure have done. This year’s data suggests that the answer is both: fashion professionals that use AI interact with it as both the engine and the steering wheel. But we also see that, for a lot of companies, both kinds of applications are still largely in the scoping or refined-pilot stage. What do you see being the trigger for deeper adoption and wider roll-out

Two things, really. The first is concrete, measurable results, seen inside a company’s own operations rather than in someone else’s case study. A pilot stays a pilot until the numbers show up on the team’s own dashboard. When peers in the same segment visibly reduce their time-to-market or their sampling costs, the internal conversation moves from “should we try this” to “why aren’t we already doing this,” and it moves fast. Adoption in this industry spreads by proof, not by persuasion.

The second is regulation, and specifically the Digital Product Passport. Mandatory from 2027, it requires brands to operate connected digital workflows that run all the way from design through to production data. Companies that have not built that infrastructure will face regulatory pressure and competitive disadvantage at the same moment, and that is what turns an interesting pilot into an operational necessity. We see that deadline as a genuine inflection point for the whole sector, and it is one of the reasons we have built the product around connected, production-grade data from the start rather than treating it as a feature to add later.