Skip to content
Train models locally: Start with notebooks and simple scriptsTrack experiments: Use MLflow to monitor model versions and metricsFocus on data quality: Clean data, handle missing valuesQuick iterations: Test different models and parametersExample Tools: Jupyter, Python, MLflow, Pandas
Basic API endpoints: Create simple REST APIsInteractive demos: Build Streamlit apps for stakeholdersReal data testing: Use production-like dataGather feedback: Test with actual usersExample Tools: Streamlit, FastAPI, Docker
Scalable systems: Handle multiple users and requestsMonitoring setup: Track model performance and driftAutomated pipelines: Set up CI/CD for model updatesPerformance optimization: Balance speed and accuracyExample Tools: Kubernetes, Cloud Services, Monitoring Tools
Prototyping
This is where ideas come to life. In this phase, UX and ML teams work in parallel to transform concepts into implementation to create testable prototypes and pipelines and deployement for to help out solution scale with demand.
Crawl-Walk-Run Approach
We start with a ✨ Minimum Viable Experience (MVE) ✨ where both teams begin simply and grow in complexity. This helps validate core concepts early and makes iteration easier.
Stage of Crawl-Walk-Run
Stage
UX Team
ML Team
Goal
Crawl
User flows
Model Developemnt
Validate core concepts
Walk
Wireframes
Pre-serving Prototype
Test key interactions
Run
High-fidelity prototypes
Scaling web to inference servers
Deliver robust solution
There are no rows in this table
🎨UX Prototypes: Growing in Detail
Each prototype builds on the previous, becoming more detailed and complex.
Crawl-Walk-Run for UX Prototypes: User Flows → Wireframes → High Fidelity Prototypes.
User Flows
What: Simple diagrams showing user journeys
Purpose: Map out key user actions, decisions, and paths
Ease of Change: Very easy - quick to edit and iterate
Wireframes
What: Basic, low-fidelity layouts showing content structure and hierarchy
Purpose: Visualize screen layouts and information architecture
Ease of Change: Moderate - requires some design work but still flexible
High Fetidity Prototypes
What: Detailed, interactive designs that look like the final product
Purpose: Test exact interactions and visual design
Ease of Change: Most complex - requires significant design effort to modify
⚙️ ML Prototypes
Sometimes called Pipelines and Deployment, this phase transforms your ML models from local experiments into production-ready systems. Before building complex serving architectures, let's make sure the core interactions work using a crawl-walk-run approach.
Crawl-Walk-Run for ML Prototypes: Model Development → Pre-Serving → Production Deployment
Model Development
# Import the ML algorithm and Iris dataset
from sklearn.ensemble import RandomForestClassifier
from sklearn.datasets import load_iris
# Load and split the classic Iris flower dataset
iris = load_iris()
X, y = iris.data, iris.target
# Train model and print accuracy
model = RandomForestClassifier().fit(X, y)
print(f"Model Accuracy: {model.score(X, y):.2%}")
Pre-Serving
Production Deployment
Pro Tip: Start simple and scale as needed. It’s easier to build complexity on a strong foundation than to fix mistakes later.
🚀 Let's Connect
I'm always excited to discuss the intersection of AI and user experience, or explore potential collaborations.
✉️
🌎 Citizenship: US, Canada
📍 Location: Toronto, ON, Canada
💼
Want to print your doc?
This is not the way.
This is not the way.
Try clicking the ··· in the right corner or using a keyboard shortcut (
CtrlP
) instead.