Articles on SourceCraft
Articles where experts from SourceCraft, Yandex Cloud, Yandex Infrastructure, and other teams cover details of SourceCraft implementation and share their development-related insights and plans going forward.
2025
Artificial intelligence in the works: exploring trends and practical applications
How do development tools transform the workflows: from classrooms to product teams?
Dmitry Ivanov, Head of SourceCraft, shared the results of the study in his article on Habr.
The article covers:
- Mass implementation of AI assistants in development.
- Transformation of labor market and requirements for developers.
- Impact on productivity of development teams.
- Willingness to invest into new technologies.
- Education of a new generation of developers.
Secret ingredients for secure development: Achieving precise and fast secret detection
Code serves as one of the primary repositories for various secrets. A single commit can accidentally expose a token or password to the public. Identifying secrets becomes a real challenge when one needs to scan all projects hosted on a popular developer platform.
Our secure development experts, Denis Makrushin, Andrey Kuleshov, and Alexey Troynikov, explained how the secret detection feature is implemented, highlighting the key criteria for secret scanning: it has to be fast, precise, and thorough.
The article covers:
- Case study analysis.
- Comparison of two modern and high-speed secret-scanning tools: Kingfisher and Gitleaks.
- Review of benefits LLMs have over traditional methods, i.e., entropy analysis, regular expressions, and signatures.
Dogfooding, AI assistants, and code navigation: Exploring SourceCraft, a new development platform by Yandex
The story of SourceCraft began in Yandex Infrastructure, a team focused on building tools for creating and deploying internal Yandex applications and services and supporting the infrastructure utilized by most of the company's engineers. This explains why most of the feature ideas for the new platform came about as a result dogfooding, i.e., using one's own product by the team of its creators.
Platform developers, Olga Lukyanova and Sergey Zakharchenko, will shed light on what it is like to create a developer platform while also using it to write code, run tests, review pull requests, and build and deploy the application.
2024
AI tool for developers: How we were training LLMs to work with code
Viktor Ploshikhin, head of ML laboratory at Yandex Platform Engineering, gave an account of his team's effort to build an AI assistant for developers. How models were fine-tuned on real code; the "whys" behind the choice to predict statements; what metrics and methods they developed for quality assessment.
What you will learn:
- What problems are addressed by the developers and what LLMs do to help.
- What modern AI assistants are capable of.
- Commonly used methods to evaluate the benefits of AI products.
- AI's impact on business.