Founded Toloka inside Yandex in 2014 and has run it as CEO since 2020, through its carve-out via Nebius and the 2025 Bezos-led round that gave it independent voting control. She also leads Mindrift, Toloka's expert-sourcing platform, so she directly owns both the lab-facing business and the expert supply side.
high confidenceNetherlandsAt Toloka since 2014Updated 2026-09-19
Background
Built Toloka as Yandex's crowdsourcing platform, serving as Yandex's Head of Crowdsourcing and Platforms (2015-2020 per Crunchbase). Took the business from mass micro-task crowdsourcing to domain experts for LLM training and agent testing after the 2023 separation from Yandex's Russian operations.
What they run now
Scaling Toloka's expert/hybrid human-AI data offer and Mindrift supply post-Bezos round; productising a self-serve platform (pipelines, exams, synthetic data) and pushing into agent evaluation and robotics data.
Career
2014 – presentFounder; CEO, Toloka — CEO title since Oct 2020 per Crunchbase
2015 – 2020Head of Crowdsourcing and Platforms, Yandex
On the record
Human experts · 2025-05
There will always be a need for human experts to control, verify and help ensure AI output quality. source
Hybrid human-AI · 2025-05
Positions Toloka on hybrid solutions: human input is essential in AI development, combined with automation. source
Growth · 2025-05
Called the 2025 investment a pivotal moment and a new phase of growth backed by investors who understand AI. source
Recent posts
5 posts archived · most engaged first, then the latest
Physical AI is also growing btw :)
Starting from the revival of good old classical crowdsourcing for large scale egocentric and audio collection, and up to UMI and Teleop manipulation data collection - we are expanding our capabilities 🦾🦾🦾
Amazon Mechanical Turk is shutting down, but it played a huge role in the industry:
the platform itself, and the community of researchers around it, have contributed greatly to the concept that human intellectual activity can be quantified, described in mathematical models, and managed programmatically at scale. It was a great source of inspiration for us to create Toloka in the first place.
Despite the fact that the complexity of human data has grown by orders of magnitude over the last 10 years (even if measured objectively by average handling time and the market price of these man-hours of work), there are still tasks that require mass, diverse, and subjective human judgments:
side-by-side design comparisons
crowd testing
data collection (for example, egocentric videos for Physical AI)
Move your crowdsourcing projects from MTurk to Toloka smoothly and get access to the power of a global crowd, plus modern techniques for agentic project creation, LLM and human QA, and highly skilled domain experts on top of the crowd workforce.
Okay, moving back from large scale crowdsourcing to highly specialised professionals in knowledge work, this time - 🎓 Legal 🎓
Next #Tolokacoolproject is about creating a benchmark for Legal RAG Challenge for Machines Can See Summit.
The team of RAG Challenge used our self-service platform to work directly with vetted legal experts, building a dataset that reflects genuine legal complexity: real regulations, case law, and contracts — not synthetic approximations.
What made it work:
— A platform built for fast iteration and quick onboarding
— Access to real domain experts, not generalist annotators
— Full control over the dataset creation process
This week's #Tolokacoolproject is about training Chain of Thought capabilities.
In order to train a model to analyse complex graphs and charts, a project was launched on @TolokaAI self-service platform that involved:
1. Sourcing diverse and properly licensed charts
2. Creating realistic and challenging prompts to describe these charts
3. Providing step-by-step human guidance on how to find answers to these prompts in charts
The secret sauce is in:
- Careful specification of quality rubrics that allow LLM QA to control the quality and automatically accept only data items that meet all quality criteria
- Access to STEM-profiled experts whose personal expertise helps generate diverse and realistic prompts
New #Tolokacoolproject will be about 📹 🤳 Egocentric Data Collection 🤳 📹
Who would have guessed!
Ironically, in the world overfed with AI-slop video, good old natural human data collection is on the rise again!
We have even integrated direct access to our global crowd of tolokers (that has been largely deprioritised compared to PhD-level domain experts in recent years), and since the opening of @TolokaAI platform in a self-service mode, cases of crowdsourced data collection are one of the most popular among our requesters.
Accounts whose posts Olga has reposted recently: Toloka, Justin Anderson, Vitaly Moiseev.
Why it matters here
Toloka's About page lists image, video and audio generation data among its specialties, and Mindrift recruits creative writers, so creative data is in scope. Toloka is also Europe-based (Amsterdam), close to Julian. The company runs a self-serve platform a small specialist could use or supply into. Moderate relevance as a potential partner or subcontracting channel.
How to reach
Public on X (@OlgaMegorskaya) and a regular conference speaker (e.g., Web Summit). Speculative: European proximity and a narrow offer (vetted design/visual-judgement experts plus rubrics that plug into Toloka's platform or Mindrift) is the most credible angle.
What we could not establish
Education not verified
Exact base city (Crunchbase location string is garbled)
Miju Labs
