Scaling Conversational AI in Production
summary
Contact center systems handle millions of conversations per day and this explains a concrete MLOps stack: a feature store for context, synthetic data generation for rare cases, experiment tracking and models registered with metadata, containerized CI/CD that cuts deployments from weeks to hours, and gradual rollout via shadow testing and canary releases. Monitoring runs on four levels (technical, data quality, performance, business), drift detection triggers automated retraining from agent-verified interactions, and rollbacks are mandatory.
AI-written summary · about
chapters (7)
- 0:00Introduction and Overview
- 0:45Challenges in Contact Center AI
- 3:04ML Ops: The Backbone of AI
- 3:47Feature Engineering and Training
- 5:50Deployment and Monitoring
- 8:29Handling Data Drift
- 9:47Conclusion and Key Takeaways
related talks
53mExperience with an end-to-end MLOps pipeline: Practical insights and 3D architectural perspectives
32mHarnessing Large Language Models for Enhanced Customer Experience
43mNavigating the MLOps Journey: Key Considerations for Successful Implementations
31mMLOps for High-Stakes Environments
47mMachine Learning in Production: An Intro to MLOps.
32mLiveMLP: ML Platform for Assisting Contact Center Agents in Real-Time
10mAvoiding the Oops with MLOps
41mTHE STATE OF MLOPS - MACHINE LEARNING IN PRODUCTION AT ENTERPRISE SCALE