Miju Labs

← All buyers · Recraft

Anna Veronika Dorogush

Founder and CEO, Recraft · Recraft

medium confidenceTier A · Talk to firstImage, video & audio modelsimage

Founded and runs Recraft, the London image/vector lab that markets its models explicitly on 'design taste'. She personally authored the V4 launch post (17 Feb 2026) stating the model was developed in close collaboration with designers and tuned around design aesthetics and professional expectations — a taste claim with no published methodology behind it. At a company this size she is the economic buyer for any data spend.

Background

Machine-learning scientist before founder: she co-created CatBoost, the gradient-boosting library, at Yandex, which is an unusual pedigree for a design-AI CEO and shows up in how Recraft talks about itself — model scale, training precision, benchmark position. She founded Recraft in 2022 and took it public on Product Hunt in May 2023 with vector generation.

Recraft raised a $12M Series A (Jan 2024) and a $30M Series B led by Accel (May 2025) with 4M+ registered users, and trains its own ~20B-parameter models from scratch rather than fine-tuning. V3 (Oct 2024) topped the Artificial Analysis text-to-image leaderboard; V4 (Feb 2026) was a ground-up rebuild aimed at prompt accuracy and design workflows.

What they run now

Recraft V4 (standard and Pro), trained from scratch on Nebius infrastructure; vector/SVG generation, brand consistency, text fidelity and layout control for professional designers and marketers. The company positions against Nano Banana Pro and GPT-class models on design usability rather than photorealism, and has 350M+ images generated on-platform as an implicit preference signal. Recraft is hiring in London, and has advertised a Data Annotation Manager role (Engineering, London) running annotation projects end-to-end on MTurk, Toloka and Clickworker.

What they buy

Part evidenced, part inference. Evidenced: Recraft runs crowdsourced annotation — its own Data Annotation Manager job specifies managing annotation projects end-to-end on MTurk, Toloka and Clickworker, writing annotation instructions and training materials, and running QC. Inference: the 'close collaboration with designers' behind V4 implies some designer panel or contractor arrangement that is not described anywhere public, and a taste-positioned model needs preference pairs and aesthetic rubrics, not just captions.

Career

  1. 2022 – presentFounder and CEO, Recraft
  2. —Machine learning — co-creator of CatBoost, Yandex

On the record

design taste as the product · 2026-02

Positions V4 as design taste meeting image generation — developed in close collaboration with designers and tuned around design aesthetics and professional expectations. source

Recent posts

We started @recraftai in 2022 to bring AI into the world of design and visuals. 3 million designers, marketers, and entrepreneurs later, Recraft is in the @TechNation Future Fifty 2026 Cohort as a graphic design research lab. It's a huge honor to be amongst other bright startups working on gene therapy, banking infrastructure, energy data, and more. The Future Fifty has supported over 35% of the UK's unicorns to date. Can't wait to see where our team goes from here! #FutureFifty #ItTakesATechNation
We redesigned Recraft Studio. The prompt menu is back on the left. That's where it was originally, but at some point we moved it to the bottom because everyone else does it that way. Bad call! The prompt at the bottom eats valuable workspace. We also reworked references and how you view the prompt behind a generated image. Users get used to how things are, but you have to keep rethinking the design so new capabilities are easy to use and the product doesn't get more complex. And sometimes you just admit you were wrong. That's the only way to build the best experience. Would love your feedback!
This one I'm especially happy about. We could already generate SVGs, but now you can edit them precisely, right on the canvas in Recraft Studio. One of the most requested features from our community. Now the whole workflow stays in one place. Let me know what you think!
Today the @MidnightCodeCup final begins in Belgrade. 24 hours, teams of three, the strongest competitive programmers in the world. We started this contest because we believe in the competitive programming community and want to invest in it. These are some of the sharpest minds out there. The format is modern, built for the AI era: use any AI you want, and the problems will still be hard. Writing problems like this is genuinely difficult. Opening ceremony in an hour. May the strongest win!

Interviews & talks

Why it matters here

High. Recraft is the purest 'design taste' buyer in the batch and the only one with a published in-house annotation operation running on generic crowd platforms. That combination is the ideal sales setup: they already pay for annotation, and MTurk/Toloka/Clickworker crowds cannot supply professional design judgement, which is exactly what their marketing claims. The gap between the claim and the supply chain is the pitch. London HQ also makes this the most time-zone-convenient target for Julian in Germany.

How to reach

Speculation: go straight at the credibility gap, politely. V4's launch claims designer collaboration but publishes no panel size, recruitment method or agreement statistics — while Ideogram publishes a designer study with numbers. Offer a professional-designer preference panel that would let Recraft publish comparable evidence for V5, and note that crowd platforms cannot staff it. She is technical and ex-CatBoost: bring method and numbers, not adjectives. Realistically the working-level entry is Pavel Ostyakov.

What we could not establish
  • An X handle (AnnaVDorogush) was recorded in the index; not independently verified in this pass, so left null here.
  • Whether the 'close collaboration with designers' behind V4 used employees, contractors or a vendor.
  • Whether the London Data Annotation Manager role has been filled, and by whom.
  • No public speaking or podcast appearances found for 2025–2026.

Why they are on the map

Founder/CEO; publicly frames V4 as 'design taste meets AI' — owns the taste positioning. source — evidence dated 2026-02

Decides whether the problem is funded.

How Recraft buys data

Recraft V4 (2026-02-17) was 'developed in close collaboration with designers, tuning it around design aesthetics and professional expectations'; no methodology or vendor disclosed. Trains own foundation models on Nebius (V4 case study, early 2026). 350M+ images generated on platform provide in-house preference signal.

CEO Anna Veronika Dorogush (ex-CatBoost lead) with Head of AI Pavel Ostyakov running ~11 reports; small Head of Design function under Ivan Olenkevich.

Sources

  1. Recraft blog: Introducing Recraft V4 (17 Feb 2026)
  2. Wikipedia: Recraft (founding, CatBoost, funding, V2/V3/V4)
  3. Recraft Data Annotation Manager job — London, Engineering (MTurk, Toloka, Clickworker)
  4. Nebius customer story: Recraft V4
  5. The Org: Recraft org chart