Track · Start with AI
AI projects for beginners: learn AI by building it
Start from your first call to a language model and work up to an assistant of your own: prompts you can test, structured output, embeddings, and automations for real inbox, document and spreadsheet work. No machine learning background is needed.
- 21
- Labs
- 17 h
- In total
- Beginner
- Level
- 1
- Free
Start the free labYour First AI Call: Talk to a Language Model from Python · 40 min
What you will build
- A Python script that calls a language model and handles keys, errors, retries and cost
- Prompts improved against a score, with a judge whose numbers you can defend
- Validated, structured output from messy files, CSVs and invoices
- Answers that quote their source, and a check that catches hallucinations
- Automations that triage an inbox, classify spreadsheet rows and draft replies behind an approval step
Before you start
- No machine learning background
- Some Python helps; the chatbot lab teaches the basics as you build
Tools you will use
PythonpandasPydanticJSONEmbeddingsGitWebhooks
Labs in this track
In order, from the first lab to the hardest. Every lab stands on its own, so start wherever you like.
1
First calls and prompts
Talk to a model from Python, see what tokens cost, and improve a prompt against a score instead of by feel.
- Lab 1Your First AI Call: Talk to a Language Model from PythonSend your first prompt to a real language model from Python, then learn the five things every AI feature is built on: reading the reply, reading the token receipt and the time it took, what temperature does to the answers, giving the model a role with a system message, and keeping your program alive when the call fails.40 minBeginnerHostedFree# ai-first-call · step 1$ lab.check(1)Step 1 Completegrade ........... pass
- Lab 2Learn Python by Building a ChatbotLearn the Python you need for AI work by building a real chat assistant, one idea per step: variables and f-strings for its instructions, lists and dictionaries for its memory, a while loop that reads what you type, if/elif/else for commands and limits, and slicing to keep a long conversation from growing without end.50 minBeginnerHostedPro# python-chatbot · step 1$ lab.check(1)Step 1 Completegrade ........... pass
- Lab 3Tokens, Context and Cost: Measure Before You SpendCount tokens with tiktoken and see what a token really is, compare your count with the provider's billed prompt tokens, price a 20,000-review job against a budget before running it, truncate text and chat history to fit a context window, and catch replies that were silently cut off by max_tokens.45 minBeginnerHostedPro# tokens-context-cost · step 1$ lab.check(1)Step 1 Completegrade ........... pass
- Lab 4Prompting That Passes Tests: Improve a Prompt with a Score, Not a HunchImprove a real support-triage prompt against a 20-case test set: state an exact output format, write the policy as rules, fence off customer text against prompt injection, add few-shot examples aimed at the cases that still fail, and confirm the result on a held-out set you never tuned on. No coding: you edit one prompt file and a harness scores it on a live model.45 minBeginnerHostedPro# prompting-that-passes-tests · step 1$ lab.check(1)Step 1 Completegrade ........... pass
- Lab 5Judge the Output: Pick the Better Prompt with Numbers You Can DefendBuild an evaluation the way AI teams do: a stratified test set with the edge cases on purpose, rule checks that need no model, an LLM judge measured against human labels before it is trusted, a pairwise comparison that cancels position bias by swapping the order, and a paired bootstrap interval that says whether the new prompt is really better or you need more data.60 minBeginnerHostedPro# judge-the-output · step 1$ lab.check(1)Step 1 Completegrade ........... pass
- Lab 6Compare AI Models Before You Trust One: Accuracy, Cost and a CascadeChoose between a small and a large language model on numbers instead of reputation, without code: measure accuracy, cost and speed on labelled return requests, repeat the runs to see how much results wobble, turn a manager's memo into a measurable bar, find where the cheap model is reliable, and build a cascade that meets both the accuracy and the budget, confirmed on next month's requests.40 minBeginnerHostedPro# compare-models · step 1$ lab.check(1)Step 1 Completegrade ........... pass
2
Work with your own data
Files, CSVs, web APIs and embeddings, with output you can validate and answers that quote their source.
- Lab 7Files and JSON with AI: Turn a Folder of Reviews into DataRead a folder of customer reviews with Python's pathlib, summarise each one with a language model, ask the model for JSON and parse it safely even when it adds fences or chatter, save and reload results as JSON, and process the whole folder so one empty or broken file is recorded instead of stopping the run.50 minBeginnerHostedPro# files-and-json · step 1$ lab.check(1)Step 1 Completegrade ........... pass
- Lab 8pandas + LLM on a Real CSV: Clean, Classify, Check, ReportTake a messy helpdesk export from raw file to trustworthy numbers: load it as text to see the real problems, clean duplicates, blank rows, spellings, three date formats and money in three styles with pandas, label 152 tickets with a language model in numbered batches that cannot silently misalign, spot-check the labels against a hand-labelled sample with a confusion matrix, and answer the manager's questions with groupby.60 minBeginnerHostedPro# pandas-llm-csv · step 1$ lab.check(1)Step 1 Completegrade ........... pass
- Lab 9Structured Output with Pydantic: Validate, Retry, GroundTurn messy supplier emails into validated Python objects: declare the order shape with Pydantic types and constraints, add an ISBN check-digit validator and a computed total, hand the model the JSON schema as response_format, retry with the validation errors when a reply is wrong, and reject valid-looking output whose values are not in the source email.55 minBeginnerHostedPro# structured-output-pydantic · step 1$ lab.check(1)Step 1 Completegrade ........... pass
- Lab 10Embeddings You Can See: Search, Map and Clean a Catalogue with VectorsTurn 200 book blurbs into embedding vectors and see what the numbers do: measure closeness with cosine similarity, find nearest neighbours, search by meaning with query and passage embeddings, draw the catalogue as a 2D map with PCA, recover the shop's sections with KMeans, and find the rows that are not books by their lack of neighbours.50 minBeginnerHostedPro# embeddings-you-can-see · step 1$ lab.check(1)Step 1 Completegrade ........... pass
- Lab 11Catch a Hallucination: Ground, Quote and Audit Every AnswerWatch a language model invent a shop's refund window, delivery prices and discounts with total confidence, then stop it layer by layer: answer only from the policy with an exact fallback for questions it does not cover, attach a verbatim quote that code verifies, split answers into claims a stronger model checks one by one, and send customers only the answers that pass every gate.55 minBeginnerHostedPro
- Lab 12Call Web APIs Safely: Keys, Errors, Retries and Rate LimitsCall a real-behaving web API from Python with requests the way production code does: keep the key in the environment and out of git, send parameters and headers with a timeout, treat 404, 401 and silence differently, retry only what can succeed with exponential backoff and Retry-After, and page through results under a rate limit before handing them to a language model and checking its answer against the data.60 minBeginnerHostedPro# web-apis-safely · step 1$ lab.check(1)Step 1 Completegrade ........... pass
- Lab 13Terminal and Git for AI Work: Review and Undo What an AI Assistant ChangesLearn the terminal and git by supervising an AI coding assistant: move around and run scripts, put a project under version control with a proper .gitignore, read the assistant's edit as a diff, keep the change you asked for while restoring the tests and behaviour it quietly broke, and undo a bad auto-committed change with git revert.45 minBeginnerHostedPro# terminal-and-git · step 1$ lab.check(1)Step 1 Completegrade ........... pass
3
Automate real work
Inbox triage, invoices into a spreadsheet, meeting notes into action items, and an assistant your team can use.
- Lab 14Email Triage with AI: Route an Inbox by What Mistakes CostTurn a shared office inbox over to a language model without writing code: write a prompt that routes every email to the right team, flags what cannot wait until tomorrow, sends ownerless emails to a person, and fixes the most expensive mistakes first, scored by a harness that prices each mistake the way the business feels it, then confirmed on emails the prompt has never seen.45 minBeginnerHostedPro# email-triage-prompt · step 1$ lab.check(1)Step 1 Completegrade ........... pass
- Lab 15Messy Documents to a Clean Table: Invoices into a Spreadsheet with AIGet a language model to fill an accounts spreadsheet from invoices in every layout, without code: name the columns, force one format per column including UK dates and European decimals, make it leave cells blank rather than invent them, write rules for the traps real invoices contain (balances brought forward, trade discounts, dotted dates), and confirm the result cell by cell on invoices it has never seen.45 minBeginnerHostedPro# docs-to-table · step 1$ lab.check(1)Step 1 Completegrade ........... pass
- Lab 16Classify 500 Spreadsheet Rows with AI and Measure the Error RateSort a real-looking customer survey spreadsheet with a language model and no code: write a category prompt a program can read, batch 500 rows into a few calls, measure the error rate honestly with a blind hand-checked sample and its confidence range, fix the overlaps with the manager's house rules, and confirm the result on next month's survey.50 minBeginnerHostedPro# classify-spreadsheet-rows · step 1$ lab.check(1)Step 1 Completegrade ........... pass
- Lab 17Meeting Notes to Action Items with AI: Owners, Dates and Nothing InventedTurn messy meeting notes into a clean action tracker with a language model and no code: a line format a program can import, owner rules that leave unowned work visibly unassigned, real due dates from a calendar the model reads instead of computes, and a list of what is not an action, scored by what each mistake costs and confirmed on meetings the prompt has never seen.45 minBeginnerHostedPro# meeting-action-items · step 1$ lab.check(1)Step 1 Completegrade ........... pass
- Lab 18Multi-Step AI Workflow: Draft, Review and Revise Without CodeBuild an AI workflow in a YAML file, with no code: split one prompt into a facts step and a writing step, give the writer rules it can follow, add a stronger model that reviews each draft for invented claims, revise only the drafts it flags, then swap in a cheaper writer and prove the workflow on books it has never seen, within a cost budget.50 minBeginnerHostedPro# multi-step-workflow · step 1$ lab.check(1)Step 1 Completegrade ........... pass
- Lab 19Build a Prompt Template Library for Your TeamTurn a team's copied-around prompts into one shared prompt library, without code: templates with declared variables, a shared house-style partial so rules live in one place, customer text fenced off against prompt injection, and an optional channel variable with a default, all run against the team's test cases and new ones the library has never seen.45 minBeginnerHostedPro# prompt-template-library · step 1$ lab.check(1)Step 1 Completegrade ........... pass
- Lab 20Build Your Own AI Assistant: A System Prompt and a Knowledge File That Hold UpBuild a customer-facing AI assistant from two text files, without code: a knowledge file curated from a messy staff handbook with every secret left out, then system prompt rules for unknown answers, scope, false assumptions, refunds, staff-only information and takeover attempts, tested on tricky questions a judge model grades and confirmed on ten it has never seen.50 minBeginnerHostedPro# your-own-assistant · step 1$ lab.check(1)Step 1 Completegrade ........... pass
- Lab 21Trigger-Based AI Automation: Webhook Inbox, Drafted Replies and an Approval Step, Without CodeBuild an automation in a YAML file that answers a vet clinic's incoming messages: ignore machine mail so it never loops with an autoresponder, classify messages and spot emergencies, draft replies from the clinic's facts, send only routine answers and queue the rest for approval or page the vet, and handle webhook re-deliveries, proven on messages it has never seen.50 minBeginnerHostedPro# trigger-automation · step 1$ lab.check(1)Step 1 Completegrade ........... pass
Graded project
Build a structured-output extraction service
Turn messy text into validated, typed JSON with a language model, then submit it for a score on every rubric criterion.
Guides for this track
Questions about this track
Some Python helps. The chatbot lab teaches the basics as you build, and several automation labs need no code at all.
No. Every lab in this track runs in your browser and the model calls are handled for you. There is nothing to install and no keys to manage.
Most take 40 to 60 minutes and save your progress, so you can finish one in a sitting or stop between steps.
Lab 1, Your First AI Call, is free with an account. The rest of the track is part of Pro, which opens every lab and practice test.
Other tracks
AI agents and MCP
Tool loops, ReAct, MCP servers and clients, multi-agent supervisors and agent evaluation.
RAG and search
Chunking, hybrid search and reranking, permission-aware retrieval, GraphRAG and RAG evaluation.
Machine learning and deep learning
scikit-learn and PyTorch on real datasets, with the metrics and debugging that decide what ships.
Every lab with Pro
This track and every other one, plus every practice test. $29.99 a month, cancel any time.