Research Manager vs Business Manager: How to Manage R&D by Alexey Gusakov (Yandex) (Category Management)

Interesting. report Management of Development and Research (Research & Development or RnD). The report is told by Alexey Gusakov, technical director of Yandex Search, who is responsible for all search technologies of the company, the development of large language models and the introduction of neural networks in Yandex services for the Russian and international markets. I am personally impressed by the fact that Alexey tells me that it is important to find a balance between science and business, so as not to become "shit" or forget about innovation.

The main points are about this Alexey talks about his role as a manager, where he must find a balance between science and business. There are two key challenges: to maintain technological leadership and to provide business with the necessary technologies, many of which are tied to ML. ML itself can be divided into three parts: product and specific orders from them, R&D and the zone of the unknown. -- Product incremental improvement due to some request for improvements from the business, there are clear metrics and generally comfortable here incremental work teams. It is important that there are challenges for the development of teams and technologies. But this incremental approach can lead to a state where progress slows. -- RnD These are technologies in which there is already investment, but the result is expected in the future. Results are not guaranteed, but likely. It is important to maintain a balance between business and RnD. It is also quite difficult to motivate teams for such projects, since people do not like to work with unpredictable results. -- Zone unknown These are technologies that are developing in the world and on which whitepapers are written. Here you can follow what is happening, read whitepapers, but you will not be able to invest much, since the zone of the unknown is very large. It is difficult to evaluate RnD activities - usually it turns out to be done only after the fact and with a large lag. It is important to conduct retrospective analysis to help determine which decisions were right or wrong. An example of a successful answer to this question is the combination of three technologies. (Speech recognition, speech synthesis and machine translation) to create synergy.

• Evaluate work in the field zone There are too many scientific articles in the world, not to read and not to try. Alexey talks about two big trends in ML: investing in reinforcement learning and passing Atari Games and investing in LLM. The first group did not invest, and the second was actively engaged. Therefore, Yandex did not start chasing OpenAI from scratch.

The presentation is finalized with a thesis about the culture of Yandex, which unites engineering and product cultures, which contributes to the development of machine learning. The balance between research and implementation is partly due to Yandex’s shared culture and workshops where scientific articles and practical applications are discussed.

By the way, the Q&A section was so interesting that I took it apart. next post

P.S. I also used to approach the projectile and performer at Techlead Conf with a reportHow RnD Appears in Large IT Companies" There, I talked about how Google, Amazon, Yandex approach to RnD looks to me and talked a little about our approach at T-Bank. It's cool that this reportResearch Manager vs Business Manager: How to Manage R&DAlexey talked about how it works from the inside of Yandex:)

#Management #Architecture #Culture #RnD