A clear place to begin
Start here directly, or take AI 101 first. No coding required; optional examples use simple arithmetic.
- Identify features, labels and sensible dataset splits.
- Explain training, loss, overfitting and common evaluation measures.
- Plan a reproducible model experiment around real operating conditions.
- Read or listen
- Explore a HerbDev example
- Try a small exercise
- Check, learn and retry
Learning is the point
There is no paywall, sign-up form or required purchase. All six lessons are available from the start. Use the scores to notice gaps in your understanding, then revisit an explanation and try again.
Vocabulary words have definitions on hover or keyboard focus, and links open their vocabulary entries. Every lesson includes a worked exercise, an answer guide and text alongside its audio.
Put it into practice · About 30 minutes
Plan and evaluate a small review classifier
Write a short experiment plan for a hypothetical classifier that flags images for human review. No coding, dataset download or paid tool is required. Your deliverable is a task definition, a data-splitting plan, a calculation sheet and a decision about what to test next.
Listen to the project brief
Read the project audio transcript
ML 101. Final practical project. Plan and evaluate a small review classifier. Write a short experiment plan for a hypothetical classifier that flags images for human review. No coding, dataset download or paid tool is required. Your deliverable is a task definition, a data-splitting plan, a calculation sheet and a decision about what to test next. HerbDev example. Use OrganicVision to think about how image analysis fits into a mobile workflow. The review classifier and counts below are invented teaching examples. They are not OrganicVision performance results. Your checklist. Define “needs review” as the positive class and describe how a human would assign that label. Keep related images from the same capture session together when splitting training, validation and test data. Write down an always-negative baseline and why accuracy alone could make it look useful. Calculate accuracy, precision and recall for 8 true positives, 4 false positives, 2 false negatives and 86 true negatives. Choose the more costly mistake for your hypothetical workflow and explain how that affects your metric choice. Specify one next experiment using validation data, keeping the test set out of repeated tuning. Pause the audio and work through the checklist. A worked solution is available on the course page. Could a teammate reproduce your calculations and explain why you chose that metric? Check the worked solution after you try. This planning exercise is separate from the scored checkpoints and is not a production-readiness certificate.
- Define the label and baseline
- Split independent examples
- Evaluate a held-out test set
- Choose the next experiment
Use the checklist to review your work. Saving uses the same cookie preference as course scores.
Compare with a worked solution
Task, split and baseline
The positive label means a reviewer should inspect the image; it is not a medical or safety diagnosis. Split by capture session before fitting any preprocessing to the training set. With 10 positive cases among 100 examples, an always-negative baseline has 90% accuracy and 0% recall. It misses every case that needs review.
Check the arithmetic
Accuracy is (8 + 86) / 100 = 94%. Precision is 8 / (8 + 4), about 67%. Recall is 8 / (8 + 2) = 80%. The classifier catches eight of ten positive cases, misses two, and sends four negative cases for unnecessary review. All four counts matter when comparing it with the baseline.
Choose a defensible next step
If missed positive cases are more costly than extra reviews, investigate whether a lower decision threshold improves recall on validation data while keeping review volume manageable. Record the threshold, metric and tradeoff before the final test evaluation. A single small test set does not establish reliable production performance; examine errors and gather representative independent examples.
Could a teammate reproduce your calculations and explain why you chose that metric? Check the worked solution after you try. This planning exercise is separate from the scored checkpoints and is not a production-readiness certificate.
Choose your next learning step →Review missed questions
Revisit questions missed on your latest checkpoint attempts. These short reviews help you practise; they do not change your best checkpoint scores. Return to the lesson checkpoint when you are ready.
Check a lesson’s answers to build your review list. Saved reviews are available after cookie acceptance.
Course complete. Keep the questions coming.
You have answered every checkpoint correctly. Use your worked exercises to explain the ideas to someone else, then apply them to a small project.
Apply it in the final project · Choose what to learn next →Keep exploring: What next? · AI vocabulary · AI 101 · Learning hub