Machine Learning Interactive System Designing
Build a machine learning system design
Follow a model from raw data to training, into production, and back again. Tick what your system needs and the diagram adds the parts. Click any box to see what it does.
Click a box, or start the walkthrough to see each stage in order.
How to use the builder
- Tick the capabilities your system needs.
- Tick the qualities that matter most.
- Click any box in the diagram to see what that part does and why it was added.
Or press Walk through the ML life cycle to step through each stage in order.
The machine learning life cycle
- Data: collect raw data, label it, and clean it.
- Features: turn raw data into model inputs. A shared feature store keeps training and live features the same.
- Training: train models, tune hyperparameters and record each experiment so you can compare and rebuild them.
- Registry: store approved model versions.
- Serving: make predictions, either in batches or live through an API, and roll out new models safely with A/B tests or canary releases.
- Monitoring: watch accuracy and data drift, and retrain when the model gets worse.
Frequently asked questions
Do I need every part?
No. A small project may only need data, training and a simple API. Add parts such as a feature store or automatic retraining when you have several models or changing data.
What is data drift?
When the data the model sees in production slowly changes from the data it was trained on. Accuracy then falls, even though the code has not changed.
See also Monitoring LLM and Epoch in model training.
5 Responses