Build Fast with AI logo
RoadmapProjectsCareer PathsCheckpoints
Start Level 0
Learning path
★Roadmap overview0Environment & Foundations1LLM Fundamentals & APIs2Building Simple GenAI Applications3RAG Systems & Intelligent Agents4Production Systems & Deployment5Advanced GenAI Techniques6Choose Your Specialization Path
Resources
Projects & notebooksCareer pathsCheckpoint tasks
  1. Home
  2. GenAI Roadmap
  3. Checkpoint tasks
Practice guide

GenAI Roadmap Checkpoint Tasks

All 35 checkpoint tasks from Levels 0 to 5 in one place. Each one is time-boxed and has concrete steps, so you know when you're done and when you're stuck.

By the Build Fast with AI team·Published April 10, 2026·Updated November 19, 2026

Checkpoint tasks
35
Levels covered
0 to 5
Per task
30 min – 6 hours
Suggested pace
8 hours a week
On this page
  1. How to use the checkpoints
  2. Level 0 tasks
  3. Level 1 tasks
  4. Level 2 tasks
  5. Level 3 tasks
  6. Level 4 tasks
  7. Level 5 tasks
  8. Official documentation
  9. FAQ

Start here

How to use the checkpoints

  1. 1

    Do them in order

    Each task assumes the ones before it, and later levels reuse what you built.

  2. 2

    Respect the time box

    If a 1-hour task takes you 4, go back to that level's topics before pushing on.

  3. 3

    Keep about 8 hours a week

    Steady weekly practice beats one long weekend a month.

  4. 4

    Commit everything

    Push each task to GitHub with a short README; together they become your portfolio.

Level 0 · 5 tasks · ~4h

Environment & Foundations

  1. 1

    Environment setup

    30 min
    • Create a genai-learning folder
    • Create and activate a virtual environment
    • Confirm python --version prints 3.10 or newer

    Time: 30 min

  2. 2

    Package installation

    30 min
    • Install requests, python-dotenv and jupyter
    • Freeze them into requirements.txt
    • Rebuild a fresh environment from that file to prove it works

    Time: 30 min

  3. 3

    API key setup

    45 min
    • Create an API key with OpenAI or Google AI Studio
    • Store it in a .env file and add .env to .gitignore
    • Load it in a script without printing it

    Time: 45 min

  4. 4

    First notebook

    1 hour
    • Launch Jupyter and create hello.ipynb
    • Mix code cells and markdown cells
    • Import and use one library in the notebook

    Time: 1 hour

  5. 5

    First LLM API call

    1 hour
    • Load the key from .env inside a notebook
    • Send one prompt and print the reply
    • Clear outputs before committing so the key never lands in Git

    Time: 1 hour

Open the full Level 0 guide →

Level 1 · 5 tasks · ~15h

LLM Fundamentals & APIs

  1. 1

    Make your first API call

    1 hour
    • Pick a provider: OpenAI, Anthropic or Google
    • Send a prompt from a Python script and print the reply
    • Print the tokens used and the estimated cost
    • Commit to Git with the key in .env, never in code

    Time: 1 hour

  2. 2

    Prompt engineering experiments

    1.5 hours
    • Write five different prompts for the same task
    • Compare the outputs side by side
    • Note which technique helped most: specific instructions, examples, a role or an output format

    Time: 1.5 hours

  3. 3

    Token counting

    1 hour
    • Count tokens for three texts with tiktoken
    • Work out what each would cost to send
    • Look up the context window of three different models

    Time: 1 hour

  4. 4

    Streaming

    1 hour
    • Stream a response token by token
    • Compare how fast it feels against a non-streaming call

    Time: 1 hour

  5. 5

    Model comparison

    1.5 hours
    • Send the same prompt to two or three models
    • Compare quality, speed and cost in a table
    • Write down which model fits which job

    Time: 1.5 hours

Open the full Level 1 guide →

Level 2 · 6 tasks · ~20h

Building Simple GenAI Applications

  1. 1

    Multi-turn chat chain

    1.5 hours
    • Build a chat loop with LangChain
    • Keep conversation history between turns
    • Test it over at least three back-and-forth exchanges

    Time: 1.5 hours

  2. 2

    Prompt templates

    1 hour
    • Create three ChatPromptTemplate templates for different jobs
    • Fill them with different variables
    • Compare the output quality of each

    Time: 1 hour

  3. 3

    Structured output

    1.5 hours
    • Define a Pydantic model for the answer
    • Get the model to return valid JSON for it
    • Parse three different inputs into Python objects

    Time: 1.5 hours

  4. 4

    Streamlit app

    2 hours
    • Build a Streamlit app with a title, an input, a response area and at least one control
    • Run it locally with streamlit run app.py

    Time: 2 hours

  5. 5

    Deploy it

    1 hour
    • Push the app to GitHub
    • Deploy to Streamlit Community Cloud or Hugging Face Spaces
    • Put the live URL in your README

    Time: 1 hour

  6. 6

    Full application build

    3 hours
    • Pick a use case: a summarizer, a tutor or a support helper
    • Add error handling and user settings
    • Deploy and document it

    Time: 3 hours

Open the full Level 2 guide →

Level 3 · 6 tasks · ~25h

RAG Systems & Intelligent Agents

  1. 1

    Basic RAG system

    2 hours
    • Load and chunk a set of documents
    • Embed the chunks and store them in a vector database
    • Answer questions using the retrieved chunks, with sources cited

    Time: 2 hours

  2. 2

    Vector database comparison

    1.5 hours
    • Load the same data into two vector databases
    • Compare query speed, metadata filtering and setup effort

    Time: 1.5 hours

  3. 3

    Agent with tools

    2 hours
    • Give an agent two tools, for example web search and a calculator
    • Log every tool call and its result

    Time: 2 hours

  4. 4

    Better retrieval

    2 hours
    • Try two chunk sizes and add a reranker
    • Measure answer accuracy on 10 test questions before and after

    Time: 2 hours

  5. 5

    Multi-agent system

    3 hours
    • Split one task across two or three agents with handoffs
    • Trace the full run with Opik or AgentOps

    Time: 3 hours

  6. 6

    RAG plus agent

    3 hours
    • Build an agent that decides when to search your documents
    • Evaluate it on a fixed question set

    Time: 3 hours

Open the full Level 3 guide →

Level 4 · 7 tasks · ~30h

Production Systems & Deployment

  1. 1

    FastAPI service

    2 hours
    • Expose your Level 2 or Level 3 app as a FastAPI endpoint
    • Validate requests and responses with Pydantic

    Time: 2 hours

  2. 2

    Logging and monitoring

    2 hours
    • Log every LLM call with latency, tokens and cost
    • Add tracing so you can replay a bad response

    Time: 2 hours

  3. 3

    Unit and eval tests

    2 hours
    • Unit-test the code that does not call the LLM
    • Add a Promptfoo suite that runs your key prompts on every change

    Time: 2 hours

  4. 4

    Dockerize

    1.5 hours
    • Write a Dockerfile for the service
    • Run the service and its database together with Docker Compose

    Time: 1.5 hours

  5. 5

    Deploy to the cloud

    2 hours
    • Deploy the container to a managed cloud platform
    • Add a health check and load secrets from the environment

    Time: 2 hours

  6. 6

    Cost and performance

    2 hours
    • Cache repeated calls
    • Route easy requests to a cheaper model
    • Compare cost per request before and after

    Time: 2 hours

  7. 7

    Production checklist

    3 hours
    • Add rate limiting, retries with backoff and input guardrails
    • Write a short runbook for when the model provider goes down

    Time: 3 hours

Open the full Level 4 guide →

Level 5 · 6 tasks · ~35h

Advanced GenAI Techniques

  1. 1

    Fine-tune a model

    3 hours
    • Prepare a few hundred input and output pairs
    • Fine-tune a small open model with LoRA
    • Compare it with the base model on a held-out set

    Time: 3 hours

  2. 2

    Multimodal feature

    3 hours
    • Build a feature that takes an image or an audio file as input
    • Handle the cases where the model misreads the input

    Time: 3 hours

  3. 3

    Advanced agent

    2-3 hours
    • Add a planning or reflection step to an agent
    • Cap its steps and cost per run

    Time: 2-3 hours

  4. 4

    Open-weight model

    2-3 hours
    • Run an open-weight model locally with Ollama or vLLM
    • Swap it into one of your earlier apps

    Time: 2-3 hours

  5. 5

    Model benchmarking

    2-3 hours
    • Build a 30 to 50 example test set from your own use case
    • Score three models on accuracy, latency and cost

    Time: 2-3 hours

  6. 6

    Advanced application

    5-6 hours
    • Combine fine-tuning, RAG or multimodal input in one deployed app
    • Write up what each choice cost and what it gained

    Time: 5-6 hours

Open the full Level 5 guide →

Reference

Official documentation

Keep these open while you work through the checkpoints.

  • OpenAI API docsAPI reference, models and guides
  • Anthropic Claude docsClaude API reference and guides
  • Gemini API docsGoogle AI Studio and Gemini API guides
  • LangChain docsChains, agents and integrations in Python
  • FastAPI docsThe Python web framework used in Level 4
  • Promptfoo docsPrompt evals and red-teaming

FAQ

Checkpoint questions, answered

What are checkpoint tasks?

Checkpoint tasks are short, time-boxed exercises at the end of each level that prove you can do what the level teaches. There are 35 across Levels 0 to 5, from 30 minutes to about 6 hours each.

Do I have to finish every checkpoint before moving on?

You don’t have to, but you should. If a task takes far longer than its estimate, that is the clearest sign of a gap, and it is much cheaper to fix now than three levels later.

Can I study offline?

You can read the levels and plan your work offline, but almost every checkpoint calls an LLM API, so you need an internet connection for the hands-on parts. Level 5 shows you how to run open models locally with Ollama.

Do I get a certificate for finishing the roadmap?

No. The roadmap is a free, self-paced curriculum. Your proof of skill is the portfolio you build: deployed apps, notebooks and write-ups from the checkpoints and projects.

How should I share my checkpoint work?

Push each task to GitHub with a short README covering what you built, what broke and what you would change. Link them from one portfolio repository so a hiring manager can scan everything in a few minutes.

On this page

  • How to use the checkpoints
  • Level 0 tasks
  • Level 1 tasks
  • Level 2 tasks
  • Level 3 tasks
  • Level 4 tasks
  • Level 5 tasks
  • Official documentation
  • FAQ
Build Fast with AI logoBuild Fast with AI

A free, project-based roadmap for learning generative AI, from your first API call to production systems.

Levels

  • Level 0: Environment & Foundations
  • Level 1: LLM Fundamentals & APIs
  • Level 2: Building Simple GenAI Applications
  • Level 3: RAG Systems & Intelligent Agents
  • Level 4: Production Systems & Deployment
  • Level 5: Advanced GenAI Techniques
  • Level 6: Choose Your Specialization Path

Career paths

  • AI App Developer Roadmap
  • AI Engineer Roadmap
  • ML Researcher Roadmap
  • AI Product Manager Roadmap
  • All projects and notebooks
  • Checkpoint tasks

Keep learning

  • AI concepts explained
  • AI glossary
  • AI project ideas
  • AI interview prep
  • Blog
  • Notebooks on GitHub

© 2026 Build Fast with AI. All rights reserved.

Privacy PolicyTerms