01
Data wrangling
Handle real-world data shape issues: join constraints, anonymized rows, nested JSON, and aggregation strategy.
/ml
Machine Learning
Build intelligence layers on messy data with a measurable quality bar and clear business impact.
Overview
What this track proves.
ML
This track is for people who want applied machine learning on real-world search and analytics data. You will work with embeddings, clustering, intent classification, opportunity scoring, and insight-to-action reporting, connecting numbers to decisions, not just building charts.
Pace
6-10 hrs/week
Prerequisites
Basic coding and spreadsheet comfort. No ML degree required.
Capstone
Reproducible notebook or script with insights, recommendations, and presentation outline.
Review
Async review on data handling, model approach, insight quality, and actionability.
Good fit
This is for you if...
You are curious about embeddings, clustering, classification, or predictive modeling.
You like experiments, comparisons, and figuring out why a pattern matters.
You want to produce models and analyses that lead to action, not just top-keyword lists.
Outcomes
What you can leave with.
The goal is proof of work, not passive course completion.
Wrangle messy search or analytics data with deliberate handling of joins, gaps, and nested fields.
Apply semantic methods (embeddings, clustering, or classification) to a focused problem.
Produce 3-5 non-obvious insights backed by evidence from your model or analysis.
Write practical recommendations tied to specific pages, clusters, or business actions.
Optional Anthropic courses when partner access allows. Read the FAQs. Your capstone remains the main credential proof.
Curriculum
The work, in order.
02
Embeddings and clustering
Group queries or pages by meaning and map clusters to content coverage gaps or cannibalization.
03
Intent and opportunity modeling
Classify intent beyond generic buckets and score where the biggest wins are hiding in existing data.
04
Insight to action
Connect model output to specific recommendations and define what counts as a non-obvious finding.
Capstone examples
The artifact can take a few shapes.
Performance-based opportunity modeling: rank content actions by expected business impact.
Semantic intent modeling: classify queries into deeper intent categories that predict behavior.
Semantic clustering and topic mapping: map demand vs. content coverage at the topic level.
Content vs. performance alignment: find where ranked content and query intent diverge.
Predictive modeling (stretch): forecast traffic, CTR, or rank movement from trend signals.
Correlation and signal analysis: surface which signals actually predict engagement or conversion.
Builder signal
What Builder-level work looks like.
Builder means you have shipped a reproducible ML analysis with semantic methods, non-obvious insights, and recommendations that connect data to a specific action.
Tool access
What you need to start.
The track is designed around accessible tools and clear alternatives. Use this as a practical setup check before applying.
ToolAccessAlternatives and caveats
Python environment
Required
Local Jupyter, Google Colab, or another notebook runtime
Your prototype must be reproducible from a script or notebook.
Data workspace
Required
Public datasets, sample CSV exports, BigQuery sandbox, or synthetic search analytics data
Do not share private client data. Use public or synthetic data for submissions.
ML libraries
Required
pandas, scikit-learn, sentence-transformers, HDBSCAN, or equivalent open-source stack
Pick tools that fit your time budget; depth of thinking matters more than stack polish.
Pythonpandasscikit-learnSQL or BigQueryEmbeddingsmatplotlib or plotlyJupyter or notebooks