YuMoney Built RAG Platform and AI Assistant Eugene for Corporate Wiki Search

YUMoney has built a RAG platform and AI assistant "Eugene" for searching corporate Wiki

Payment service YUMoney explained on Habr how it built its own RAG platform (Retrieval-Augmented Generation) and AI assistant "Eugene" — it replaced the standard search across corporate Wiki and understands the meaning of an employee's query, rather than simply matching keywords.

Why standard search stopped working

As companies grow, they accumulate thousands of pages of internal documentation, and finding the right information in it becomes increasingly difficult. According to YUMoney's account, standard search across corporate Wiki can no longer cope with this volume: employees waste time navigating through pages and instead of searching for answers independently, they repeat the same questions to colleagues.

• Company — YUMoney, a Russian payment service
• Name of the assistant — "Eugene", built on RAG technology
• Author of the material — Ilya, director of development at YUMoney
• Key decision by the team — rejection of off-the-shelf RAG solutions in favor of their own platform
• Target audience for the material — architects, AI engineers and specialists implementing corporate LLM solutions

How AI assistant "Eugene" works

Eugene combines corporate knowledge of the company with the capabilities of large language models. RAG stands for Retrieval-Augmented Generation — "generation with augmentation through search": first, the system finds fragments of documentation in the knowledge base that are relevant to the question, and then the language model formulates an answer based on them, rather than simply on the data it was originally trained on. This allows answering based on the meaning of the query, rather than formal matching of words, as in classical full-text search across Wiki.

According to YUMoney's team, the transition of large language models into the category of working tools made it possible to combine accumulated corporate knowledge with the capabilities of LLM — this is how Eugene emerged, capable of understanding the meaning of an employee's question, rather than just the keywords in it.

Why YUMoney rejected ready-made solutions

Ilya, director of development at YUMoney, promises readers to explain why the team rejected ready-made RAG solutions on the market and chose the path of building their own platform for corporate knowledge. The author addresses the material primarily to architects and AI engineers who face similar choices when implementing LLM within a company — that is, they decide whether to embed a ready-made product or build a platform for their corporate data independently.

YUMoney's material is presented as a practical analysis of precisely this choice: why in-house development turned out to be preferable to purchasing or connecting a third-party RAG service for searching through internal company documentation.

"My name is Ilya, I am the director of development at YUMoney.

I will tell you why we rejected ready-made solutions and how we built a RAG platform for corporate knowledge," the author writes on Habr.

What this means

YUMoney demonstrates the typical 2026 path of corporations from scattered documentation in Wiki to an internal AI assistant based on RAG — a technology that is becoming the standard for knowledge search within companies with large amounts of internal documentation, and YUMoney's material is addressed primarily to those who are themselves implementing such solutions.