Python Developer for LLM Project Development Team for B2B

Python Developer for LLM Project Development Team for B2B

We are working on a B2B platform for automated responses to questions about documentation based on LLM models and search technologies.

The system is used:

• To assist support operators of various services
• To provide answers based on general-purpose documentation for various clients, both external and internal within Yandex
• In beta mode — for some Yandex Cloud clients

We plan to expand the platform's scope of use, scale the system, and automate the tools for its application.

Tasks that await you

Development of new algorithms for data processing and generation using ML models (LLM and others) Our backends use various ML models for text generation, search, ranking, and data moderation tasks, and we continuously work to ensure the backends perform well and quickly on different data, considering the business logic requirements of various clients. To achieve this, we are rethinking approaches to GPT generation while simultaneously solving common backend tasks such as observability and fault tolerance.

Creating a universal platform We are not only building a product for external customers but also creating a technological platform within Yandex so other development teams can reuse our results. For this purpose, we plan to create a unified common platform library that can be used in other developments within Yandex. We also intend to create tools to help colleagues from the ML team analyze and debug the work of ML models and the entire ML pipeline as a whole.

System scaling tasks As the system evolves, we face increasingly complex tasks of processing large amounts of data and accelerating the system's operation: we are improving the data preparation process and optimizing the response speed of our backends on this data.

Creating an isolated version of the system We plan to enable running all system components in separate, isolated environments independent of Yandex's infrastructure. To do this, we need to analyze all system components and features and design how everything will work in the isolated version.

We expect that you

• Know classical algorithms and data structures
• Write reliable and scalable code, cover it with autotests
• Have developed backends using asynchronous Python 3
• Know from experience how to create a good API
• Have worked with databases (e.g., Postgres), know how to optimize database structure and SQL queries
• Understand the basic principles of authentication and authorization when working with web services
• Can decompose a complex task

Will be a plus

• Can design distributed and fault-tolerant systems
• Have worked with ML model infrastructure, understand the basic principles of ML or want to learn them
• Can maintain backend operation in production: analyze logs and system performance metrics, work in a Unix-like environment
• Have worked with Docker and cloud systems
• Have worked with distributed data storage and processing systems (such as Hadoop, YTsaurus, etc.)
• Have worked with vector search databases