Improving
Machine Learning

Workflows
Training, Packaging and Serving.
We build speech and NLP models 

to help call centre agents
to better assist customers in real time.
Struggles
Manual Work
Struggles
Copying data from server to
server
Training &
Experimentation
Keeping track of
experiments
Keeping track of artefacts
Reproducibility
Packaging & versioning
models
Manual deployments
No CI/CD
Longer release cycles
Slow model / code
integration
No transparency
Serving
Versioning
Model encapsulation
- Machine Learning Lifecycle.
- Building Blocks: 

mlflow & TensorFlow Serving.
- Infrastructure: 

Putting all components
together.
- CI/CD for ML models.
Agenda
Building blocks: 

mlflow & TensorFlow Serving
Tracking models with mlflow
Tracking models with mlflow
Tracking models with mlflow
Tracking models with mlflow
Tracking models with mlflow
Serving models with TensorFlow
Serving models with TensorFlow
Infrastructure
Before After
CI/CD for ML models
Tracking models with mlflow
Aigent’s ML lifecycle
Solutions
Manual Work
Training &
Experimentation
No CI/CD Serving
Wilder Rodrigues
@wilderrodrigues


medium.com/@wilder.rodrigues
- Software & Artificial Intelligence Engineer
- Apache™ Committer & PMC member
- Keras contributor
- School of AI Utrecht Dean
- Amsterdam AI Ambassador
- Public Speaker
Improving  Machine Learning
 Workflows: Training, Packaging and Serving.

Improving Machine Learning
 Workflows: Training, Packaging and Serving.