#057 ML Антонина Горячева. Как отвечать за весь ML в компании, чтобы все работало

#057 ML Антонина Горячева. Как отвечать за весь ML в компании, чтобы все работало

#057 ML Antonina Goryacheva. How to be responsible for all ML in the company so that everything works

by Machine Learning Podcast · View the channel on PSI

Applied Sciences & Engineering AI — Models & Research UYQArtificial intelligence (AI)UYQMMachine learning
🎙 Machine Learning Podcast 👥 5K 📅 October 28, 2024 ⏱ 63 min 👁 572 📄 expert opinion 🧭 2026-10-16
Available in: English (current) Français

Keywords

machine learningSberMarketML infrastructurerecommendation systemsMLOps

Summary

This podcast episode features Antonina Goryacheva, Head of ML at SberMarket, a major Russian online grocery delivery platform. After a brief introduction of her background — a degree in information-analytical security systems from MEPhI and her first job as a data analyst at Megafon — she details how she grew within SberMarket from an ML engineer to team lead and eventually to Head of ML. She outlines the three main ML business domains: operations, customer-facing services, and content. For operations, she covers order forecasting, courier and picker attrition models, ETA prediction, and surge-demand management algorithms that balance service quality and cost. For customers, she describes recommender systems, personalized search and ranking, marketing personalization, and price-elasticity models to optimize discount budgets. For content, she discusses product card categorization, automated photo processing for retouchers, and an out-of-stock model that improves order integrity. She also emphasizes the importance of unified ML infrastructure and standardized templates, the role of MLOps teams, and the need to develop leaders within her team to handle the breadth of responsibilities. The conversation closes with practical advice on skills, careers, and what makes ML teams effective in a corporate environment.

191 words

Critical Evaluation

Value of the Information & Strength of the Argument

The value of the information is high for practitioners interested in how ML is structured in a large e-commerce/delivery company. Antonina provides concrete, real-world examples (e.g., demand forecasting, ETA models, surge pricing, discount elasticity) and explains how they translate into business metrics such as cost per order and ROI. The argumentation is coherent and experience-based: she consistently links ML use cases to business outcomes, illustrating with pilot results and comparative improvements. She is honest about limitations, notably that she no longer writes code herself and relies on team leaders, and she acknowledges the challenge of keeping context across the whole company. The reasoning is pragmatic rather than theoretical, and its strength lies in the credibility of a practitioner who has lived the progression from engineer to executive.

Scientific Rigor, Source Quality, Title Accuracy

Rigor is moderate: the content is well-structured and internally consistent, but the podcast format is conversational, with no formal citations, figures, or references displayed during the discussion. The description provides links to related resources (Telegram channels, the podcast page, a DS Meetup talk by the same guest), which supports traceability of some claims. The title matches the content well, accurately framing the episode as an exploration of leading ML at company scale. No audience comments were provided, so no public reception analysis is possible.

226 words

Title / Content Match

The title accurately reflects the content: the entire episode focuses on how Antonina manages ML across the whole company to keep it running, from business domains to infrastructure and team leadership.

Quality & Reliability

7/10

The speaker is an experienced ML leader with a concrete track record at a major delivery company, and the discussion is grounded in real operational cases. However, the format is an informal conversational podcast without formal citations, verifiable data, or peer-reviewed references, so reliability relies on the guest's credibility and firsthand experience.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Cited Sources

Concurring Sources

External References

Contribution & Novelties

The episode offers a rare first-person account of what it means to lead all ML operations in a major grocery delivery company. It provides a structured breakdown of how ML is organized across three business domains (operations, customers, content), with concrete use cases and honest insights into the challenges of scaling ML in a corporate environment. The discussion of infrastructure unification and MLOps templates as a management lever is particularly valuable for practitioners. It also demystifies career progression in ML, showing a path from engineer to executive based on management skills rather than solely hardcore technical expertise.

Pour aller plus loin :

159 words

Radar Profile

The radar profile shows a balanced but moderate shape: high scores in information quantity and quality reflect a content-rich expert interview, while technical depth is moderate and reliability is constrained by the informal format without formal citations. This yields a solid, informative episode with practical value.

Reliability 7/10