Session Outline

At Bolt, we want scientists and engineers to solve customer’s problems, not infrastructure ones. That’s why we’ve been investing into centralised machine learning infrastructure for 3 years, and it helped us to scale 50+ of projects, making billions of real-time inferences weekly. In this talk I’ll share how we’ve built abstraction over Machine Learning infrastructure, enabling teams to move fast, while platform teams can develop value-adding features  and get them adopted easily.

Key Takeaways

  • Even though there are many “tools for the job” in the data and ML space, leaving choice of them to product teams creates duplication of components and expensive tech debt. 
  • We believe that for a common problem there should be a common reusable solution, providing smart defaults. 
  • With good abstraction over data and ML pipelines data infrastructure team easily migrated users to better solutions (SaaS, Cloud or internal ones), optimised costs globally across the company and solved issues for 50+ ML projects on the platform with a single release
  • Centralised platform allowed Bolt ML platform team to build and integrated value-added components such as data monitoring and no-code smoke/load with   negligible cost to client teams


Speaker Bio

Denys Kovalenko – Engineering Manager | Bolt

Denys is passionate about applying data and machine learning to solving real-world problems. Currently he is leading Machine Learning and Experimentation platform teams at Bolt, and he was one of the first engineers at the Data Platform group at Bolt.  In his free time, Denys dances bachata, rides snowboard and participates at hackathons.

November 9 @ 11:40
11:40 — 12:10 (30′)


Denys Kovalenko – Engineering Manager | Bolt