Practical AI MLOps and tracking experiments with Allegro AI
Jul 20, 2020
Nir Bar-Lev, CEO and co-founder of Allegro AI, dives into the intricacies of MLOps and its unique challenges compared to traditional DevOps. He discusses the critical importance of data versioning and experiment tracking in machine learning projects. The conversation highlights the shift from individual efforts to collaborative model training while emphasizing agile innovation and improved engineering practices. Nir also shares insights about Allegro's platform capabilities in automating ML tasks and the necessity for efficient dataset management in AI development.
AI Snips
Chapters
Transcript
Episode notes
Data Scientists and Tooling
- Data scientists, trained in a scientific paradigm, initially resist adopting rigorous tooling and processes.
- This mindset shift is changing as companies integrate data scientists into product teams, emphasizing product development over research papers.
Resistance to New Tools
- Nir Bar-Lev recounts past resistance to adopting SQL databases.
- He draws parallels to similar resistance to MLOps tools, highlighting a 'not invented here' mentality.
MLOps vs. DevOps
- MLOps focuses on managing the unique challenges of machine learning development.
- These challenges include running experiments at scale, managing changing code, and handling complex environments.