Session Name: How to solve technical debt in AI System development with DevOps?
Growing interest in development AI systems also brings some challenges besides data models, such as technical dept, deployment of the AI system timely. Statistically, more than 65% of companies are taking longer than a month to deploy a developed model. There is a huge knowledge gap in understanding how foster collaboration between data science teams and other stakeholders. The purpose of collaboration is to evolve the model and maintain the AI system relevant to a user’s need. However, there are challenges which are hidden feedback loops, configuration management complexity, data dependencies, and end-2-end development pipeline. These challenges can be overcome with common DevOps practices including continuous feedback and continuous integration and deployment. We may call it MlOps or something, but the root of the solution is DevOps.
Hasan Yasar is the Technical Director of Continuous Deployment of Capability group in Software Engineering Institute, CMU. Hasan leads an engineering group to enable, accelerate and assure Transformation at the speed of relevance by leveraging, DevSecOps, Agile, Lean AI/ML and other emerging technologies to create a Smart Software Platform/Pipeline. Hasan has more than 25 years’ experience as senior security engineer, software engineer, software architect and manager in all phases of secure software development and information modeling processes. He is also Adjunct Faculty member in CMU Heinz Collage and Institute of Software Research where he currently teaches “Software and Security” and “DevOps: Engineering for Deployment and Operations.”