AlgoCademy
Start learning free

CoursesData & AIAI Engineering

Build an AI support assistant and test it like any other code.

For people who know Python and want to build with language models. You add a support assistant to the shop. It answers from the shop’s documents, looks up orders with tools, and gets measured with evals before it ships.

Start the first lesson See the plans
You type every line A real checker runs your code Nothing to install
Lessons
129
Chapters
13
Language
Python
Level
Intermediate
Interview questions
10
Free
The first chapter
Before you start
Python: lists, dictionaries and functions.

What you build

A support assistant

It finds a customer’s orders, follows the shop’s rules and asks before it acts.

Policy Q&A with sources

Retrieval over the shop’s documents, with the source cited.

An eval suite

A test set, scores, a judge model and a check on every change.

A returns agent

It plans the steps, stays within a budget and knows when to stop.

Your own coding agent

It reads a project, edits files, runs the tests and shows the diff.

An MCP server for the shop

Used from your assistant, Claude Desktop and Claude Code.

Tools and topics

Syllabus

01Calling a Model15 lessons
  1. Your First Model Callfree
  2. System and User Messagesfree
  3. A Conversationfree
  4. Stop Reasons and max_tokensfree
  5. Tokens and Costfree
  6. Streaming Answersfree
  7. When the API Says No: 429 and 529free
  8. Timeoutsfree
  9. Choosing a Modelfree
  10. Show It a Picturefree
  11. Hand It a PDFfree
  12. Let It Thinkfree
  13. Keys Live in the Environmentfree
  14. An AI Endpoint in Flaskfree
  15. Project: Product Descriptionsfree
02Prompts That Hold Up10 lessons
  1. Clear Instructions
  2. Show Examples
  3. Ask for a Format
  4. Wrap User Input in Tags
  5. Prompt Injection
  6. Saying "I Don't Know"
  7. Prompt Templates
  8. Version Your Prompts
  9. The Assistant's System Prompt
  10. Project: Policy-bound Answers
03Structured Output11 lessons
  1. Ask for JSON
  2. Parse It Safely
  3. Validate the Fields
  4. Retry with the Error
  5. The API Enforces the Shape
  6. Classify the Message
  7. Extract the Order Number
  8. Many Messages at Once
  9. The Batch API: Half Price, Later
  10. Fall Back Gracefully
  11. Project: Triage the Inbox
04Tool Use14 lessons
  1. Describe a Tool
  2. The Model Asks for a Tool
  3. Run It and Send the Result
  4. The Tool Loop
  5. Two Tools
  6. When a Tool Fails
  7. Several Calls at Once
  8. Find a Customer's Orders
  9. Rules Inside the Tool
  10. Ask Before Acting
  11. Only Allowed Actions
  12. A Step Limit
  13. Log Every Tool Call
  14. Project: The Support Assistant
05MCP6 lessons
  1. Your Shop as an MCP Server
  2. More Tools and a Resource
  3. Use an MCP Server from Your Assistant
  4. Use It from Claude Desktop and Claude Code
  5. Serve It over HTTP
  6. Ask Before a Tool Changes Something
06Retrieval (RAG)15 lessons
  1. Why the Model Doesn't Know Your Shop
  2. Split Documents into Chunks
  3. Embeddings
  4. Similarity
  5. The Top Matches
  6. Answer from the Passages
  7. Cite the Source
  8. Nothing Relevant: Say So
  9. Chunk Size Trade-offs
  10. Keywords and Vectors Together
  11. Filter by Category
  12. Keep the Index Fresh
  13. Injection Hidden in Documents
  14. Project: Policy Q&A with Sources
  15. Store the Index in a Database
07Evals12 lessons
  1. Why Evals
  2. A Test Set
  3. Exact-Match Scores
  4. Score Tool Calls
  5. A Judge Model
  6. Catch a Regression
  7. An Eval Report
  8. Hard Cases on Purpose
  9. Cost and Speed in Evals
  10. Test Cases from Real Conversations
  11. Evals on Every Change
  12. Project: An Eval Suite for the Assistant
08Agents9 lessons
  1. What Makes an Agent
  2. Plan the Steps
  3. Budgets: Steps and Cost
  4. Knowing When to Stop
  5. A Scratchpad
  6. A Return in Several Steps
  7. When Not to Use an Agent
  8. Project: The Returns Agent
  9. Hand Off to a Specialist
09Build Your Own Coding Agent8 lessons
  1. A Coding Agent Starts: Read the Project
  2. Let It Edit a File
  3. Run the Tests
  4. Fix a Failing Test
  5. Ask Before Risky Commands
  6. Step and Output Limits
  7. Show the Diff
  8. Project: Your Own Coding Agent
10Context Engineering4 lessons
  1. When the Chat Gets Too Long
  2. Summarize Old Turns
  3. Remember the Customer Between Chats
  4. Cache the Long Prompt
11Production10 lessons
  1. Cache Repeated Questions
  2. Retries and Backoff
  3. Fall Back to Another Model
  4. Cost per Request
  5. Logs Without Personal Data
  6. Keep Card Numbers Out
  7. Stream to the Browser
  8. Guardrails on Answers
  9. Settings per Environment
  10. Project: Ship the Assistant
12Capstone5 lessons
  1. The Assistant End to End, 1: Knowledge
  2. The Assistant End to End, 2: Tools
  3. The Assistant End to End, 3: Guardrails
  4. The Assistant End to End, 4: Evals
  5. The Assistant End to End, 5: Shipping It
13Interview Questions10 lessons
  1. Find the k Closest Chunks
  2. Split a Document into Chunks
  3. Did Retrieval Find It?
  4. Score an Answer
  5. Fit the History in the Budget
  6. Text That Gives Orders
  7. Write the Tool Loop
  8. Cache Model Answers
  9. Rerank What You Retrieved
  10. Ask Three Times, Take the Majority

Questions

Something else?
team@algocademy.com

Do I need an API key?

No. You write the calls with the real Anthropic SDK, and a course model in the checker answers them. Every learner gets the same answer, and nothing is billed.

More in Data & AIMachine LearningData EngineeringSQL & DatabasesAll courses

LESSON 1

Your First Model Call

The first chapter is free, no card needed. The rest comes with PRO.

Start the lesson See the plans