Artificial Intelligence seen from the software development lifecycle perspective
summary
Argues that strong AI prototypes fail in production without infrastructure, QA and CI/CD and demonstrates practical fixes: quick demo UIs via Streamlit and Hugging Face Spaces (anonymization and Stable Diffusion demos run successfully), experiment tracking and versioned models with MLflow model registry, and end-to-end pipelines with Kubeflow or Vertex AI for automated continuous training and monitoring. Claims models degrade in real traffic unless metadata, logging, retraining triggers and latency monitoring are wired into delivery.
AI-written summary · about
chapters (5)
- 0:00Start
- 11:00Machine Learning training
- 17:12Machine Learning "Operations"
- 41:21Model cards The README.md for Al Models
- 44:06Main Conclusions
related talks
16mOperationalizing AI in Finance: Real-Time Analytics
44mAI workflows and teaming - how do teams deliver AI products?
50mRealities & Misconceptions about Enterprise Applications of AI/ML and Data
26mCompleting the machine learning loop
26mbuilding reliable AI systems in production
33mAI with a DevOps mindset - experimentation, sharing ..
59mContinuous Intelligence: Keeping your AI Application in Production
37mThe AI Factor: The Blueprint for Intelligent Deployment