Free AI Tutorials:
From Beginner to AI Engineer

1. Create a file called word_count.py. A string stores text; split() separates it into words; a dictionary stores each word and its count.
2. Run the example with python word_count.py in your terminal. The result should be {'learn': 2, 'ai': 1}.
3. Replace the sample text with your own sentence. Then move the counting logic into a function that accepts text and returns the dictionary.

text = 'Learn AI learn'
counts = {}
for word in text.lower().split():
    counts[word] = counts.get(word, 0) + 1
print(counts)

Your checkpoint
You can explain the loop and reuse the function on a new sentence. This simple example keeps punctuation attached to words; try improving that next.

1. Choose a small labeled dataset. Each row is an example, input columns are features, and the value you want to predict is the target.
2. Separate training examples from test examples before fitting preprocessing or a model. Use training data to learn patterns and untouched test data to evaluate them.
3. Write down a simple baseline, such as always predicting the most common class. Compare your model against it and inspect the mistakes, not just one score.

Your checkpoint
You can distinguish classification (predicting a category) from regression (predicting a number), and explain why testing on training examples is misleading.

1. Ask an available language model to explain a short invented term without giving a definition. Record whether it admits uncertainty or invents an answer.
2. Provide your own definition in the next prompt and ask the same question. Compare the answer with the definition you supplied.
3. Identify which claims are supported by your text. A fluent answer is not evidence of accuracy: check factual claims against reliable sources.

Your checkpoint
You can explain that tokens are units of text processed by a model and that the context window limits how much information it can consider at once.

1. Begin with a vague request such as “summarize this.” Then specify the reader, source material and desired format.
2. Try the template below with a short article you can verify. Add one example of the desired answer if the format is inconsistent.
3. Check every bullet against the source. Compare the revised prompt with the original on three different inputs instead of judging a single answer.

Task: Summarize the source for a beginner.
Source: [paste the text here]
Format: Three bullets, each under 20 words.
Use only facts in the source.
If information is missing, say it is not provided.

Your checkpoint
You have a reusable prompt and a checklist for factual accuracy, format and relevance.

1. Choose three short documents you have permission to use. Split them into passages and attach a document name and passage number to each.
2. For a question, first select the passages that actually contain evidence. Paste those passages into a prompt and request an answer with passage references. This manual prototype lets you test the workflow before adding a search index.
3. Test both answerable and unanswerable questions. Record whether the selected passages are relevant, the answer is supported and the cited passages contain the claim.

Your checkpoint
Your prototype answers from supplied evidence and says when the documents do not contain an answer. Automated RAG adds retrieval; it does not guarantee factual answers.

1. Define a tool named lookup_order with one required argument, order_id. Write down its input type and the fields it is allowed to return.
2. Use mock orders to test a valid ID, an unknown ID and a malformed argument. Validate the arguments in application code before calling the tool.
3. Add a maximum number of tool calls and a clear failure response. Require a separate explicit confirmation before any later extension can cancel or modify orders.

Your checkpoint
You can explain the difference between a model requesting a tool call and your application validating and executing it.

1. List the runtime, dependencies, start command and configuration needed to run your project. Keep credentials in server-side configuration, outside public files and source control.
2. Run the application in a clean environment. Add a health endpoint that reports readiness without revealing secrets or user data.
3. Test missing configuration, a failed upstream request and a slow response. Document how to roll back a release before making the application public.

Your checkpoint
Someone else can follow your setup instructions, start the app and diagnose a failed request.

1. Save ten representative inputs, expected properties and known failure cases. Include a question with insufficient evidence and an unexpected input format.
2. Run the same inputs before and after a prompt or model change. Record factual support, formatting errors, response time and usage where available.
3. Decide which failures block a release. Keep the previous prompt and configuration so you can revert a change that makes the application worse.

Your checkpoint
You have a repeatable comparison that shows whether a change improved your application on the cases you tested.

1. Choose a narrow problem, such as answering questions about a public product manual. Define the intended user and the questions the assistant should handle.
2. Combine the RAG prototype with a simple interface. Show source passages alongside answers and provide a useful response when information is missing.
3. Publish a README with setup instructions, example questions, evaluation results and limitations. Use sample data in public demos and remove credentials before sharing code.

Your checkpoint
Your portfolio shows a working example, reproducible setup and evidence of what the project can and cannot do.

1. Describe the user problem, your approach and the result. Include a diagram or short explanation of how data moves through your application.
2. Discuss one tradeoff, one failed approach and one measurable improvement. Avoid claiming results you did not measure.
3. Ask a peer to run your project from the README. Use the feedback to improve the instructions and choose the next skill to practice.

Your checkpoint
You can discuss your design choices and demonstrate your work. Completing this roadmap alone does not guarantee a job.