A map of the whole AI field for 2026 - what machine learning, deep learning and large language models actually are, which skills every path shares, and how to pick between building AI products, engineering AI systems or researching models.
7 stages
6-18 months, depending on path
Beginner friendly
Realistic timeline
6-18 months at 10-15 hours per week
Stages
7 in order
Starting point
No AI background needed
Artificial intelligence is not one skill. It is a set of related fields - classical machine learning, deep learning, natural language processing, computer vision and, most visibly right now, applications built on large language models. This AI roadmap explains how those pieces fit together and gives you a learning order that works whichever direction you eventually take.
Use it as your starting page. The first four stages are shared by almost everyone learning AI. After that, the roadmap branches: if you want to ship AI features inside real products, continue with the AI Developer Roadmap. If you want to own models, pipelines, evaluation and production reliability, the AI Engineer Roadmap goes much deeper.
Overview
The AI learning roadmap at a glance
Shared foundations first, then a branch that matches the work you want to do.
Select a stage to see what it covers. Filter by level to focus.
1. Programming with Python
6-10 weeks
Write small Python programs confidently, because almost every AI tool assumes you can.
Variables, functions, loops, lists and dictionaries
Reading files, handling errors and using packages
Virtual environments and installing libraries
Jupyter notebooks for exploration
Git basics and a GitHub profile
Done when: A script that reads a CSV, cleans it and prints a useful summary - committed to GitHub.
Stages one to four are common ground. Stages five to seven are where you choose a direction.
1
Foundation · Beginner
Programming with Python
6-10 weeks
Write small Python programs confidently, because almost every AI tool assumes you can.
Python is the working language of AI - not because it is the fastest, but because the libraries, tutorials and research code are written in it. You do not need to be an expert. You do need to read other people's code, install packages without panic, and debug your own mistakes. Put your code on GitHub from the first week so version control becomes a habit rather than a hurdle.
Skills
Variables, functions, loops, lists and dictionaries
Reading files, handling errors and using packages
Virtual environments and installing libraries
Jupyter notebooks for exploration
Git basics and a GitHub profile
Milestone
A script that reads a CSV, cleans it and prints a useful summary - committed to GitHub.
Project ideas
An expense summariser for a bank export
A word-frequency analyser for any text file
2
Foundation · Beginner
Data and the maths that matters
6-8 weeks
Understand data well enough to know when a model is being fed rubbish.
Most AI failures are data failures. Learn to load, clean, filter and plot data with pandas and a charting library. Alongside that, build an intuition for the maths: vectors and matrices (how data is represented), basic probability (how models express uncertainty) and the idea of a gradient (how models learn). Intuition first, formal proofs only if you head toward research.
Skills
pandas and NumPy for tabular and numeric data
Plotting distributions and spotting outliers
Vectors, matrices and dot products
Mean, variance, probability and conditional probability
What a derivative and a gradient represent
Milestone
Explore a public dataset and write up three findings with charts that support them.
Project ideas
Exploratory analysis of a public transport or weather dataset
3
Core skills · Beginner
Machine learning fundamentals
8-10 weeks
Train, evaluate and honestly judge a classical machine learning model.
Classical machine learning teaches the ideas every AI system still depends on: training versus test data, overfitting, evaluation metrics and the trade-off between a simple model and a clever one. scikit-learn is the right tool here. The goal is not to memorise algorithms but to be able to say why a model is performing well or badly.
Skills
Supervised versus unsupervised learning
Regression, classification and clustering
Train, validation and test splits; cross-validation
Accuracy, precision, recall, F1 and when each matters
Overfitting, underfitting and regularisation
Feature engineering basics
Milestone
A model that beats a sensible baseline, with a written explanation of its errors.
Project ideas
A house price or churn predictor with a baseline comparison
4
Core skills · Intermediate
Deep learning and language models
6-10 weeks
Understand neural networks and how large language models work at a conceptual level.
Neural networks power image recognition, speech and language models. Train a small network yourself in PyTorch so the ideas of layers, loss and backpropagation stop being abstract. Then learn how transformers and large language models work: tokens, context windows, embeddings, attention, and why these models can be fluent while still being wrong. This stage is what lets you use AI tools with judgement instead of faith.
Skills
Neurons, layers, activation functions and loss
Training loops in PyTorch
Convolutional networks for images (overview)
Transformers, tokens and attention
Embeddings and semantic similarity
Why language models hallucinate
Milestone
Train a small image or text classifier and explain each line of the training loop.
Project ideas
A handwritten digit classifier
A sentence similarity demo using embeddings
5
Tools · Intermediate
Working with AI models and APIs
4-6 weeks
Use hosted and open models programmatically, with sensible prompts and cost awareness.
Most AI work today starts from an existing model rather than training one. Learn to call model APIs, write structured prompts, request JSON output, and run an open model locally or through Hugging Face. This is also the stage where you learn about rate limits, token costs and keeping API keys out of your code.
Skills
Calling hosted model APIs from Python
Prompt structure, examples and output formats
Hugging Face models and pipelines
Running a small open model locally
Tokens, cost and latency trade-offs
API key and secret handling
Milestone
A small tool that uses a model API to turn messy text into validated structured data.
Project ideas
A meeting-notes summariser that outputs action items as JSON
6
Specialisation · Intermediate
Choose a direction
8-16 weeks
Commit to one path and go deep enough to build something real in it.
Here the roadmap branches. AI application development means building products with models - retrieval-augmented generation, agents, AI features in web apps; follow the AI developer learning path. AI engineering means owning the systems - evaluation, fine-tuning, pipelines, deployment and monitoring; follow the AI engineering roadmap. Data science and ML research lean harder on statistics and maths, and usually benefit from formal study.
Skills
Retrieval-augmented generation (RAG) and vector search
AI agents and tool use
Fine-tuning and model evaluation
Computer vision or speech, if that is your interest
Responsible AI: bias, privacy and safety basics
Milestone
One substantial project in your chosen direction, deployed or reproducible by someone else.
Project ideas
A question-answering app over your own documents
An evaluated fine-tune of a small model
7
Job ready · Advanced
Portfolio, deployment and interviews
6-8 weeks
Turn what you learned into evidence an employer or client can check.
Deploy at least one project so it can be used, not just read. Write short case studies that explain the problem, the approach, how you measured quality and what you would improve. Then prepare for interviews, which in AI roles usually mix coding, ML concepts and a system design conversation about an AI feature.
Three documented projects, one deployed, and a practised explanation of each.
Project ideas
Rebuild your strongest earlier project to production quality
Progression
From AI beginner to specialist
What each level looks like in practice, so you know where you stand.
Beginner
Months 0-4
You can program in Python, handle data and understand what machine learning is really doing.
Focus on
Python and Git
pandas and plotting
Core ML concepts with scikit-learn
Evaluation metrics
You can: Train and fairly evaluate a simple model.
Intermediate
Months 4-9
You understand neural networks and language models, and can build with model APIs.
Focus on
PyTorch basics
Transformers and embeddings
Prompt engineering
RAG fundamentals
You can: Build a working AI feature that uses a model sensibly.
Advanced
Months 9-18
You have specialised and can design, evaluate and ship AI systems in one area.
Focus on
Evaluation and testing of AI output
Agents or fine-tuning
Deployment and monitoring
Cost and safety trade-offs
You can: Own an AI project end to end and defend its design.
Technologies
Technologies across the AI landscape
Dark chips are worth learning early. Light chips can wait until a project needs them.
Programming and data
The base layer for every AI path.
Start with
Python
Git
pandas
NumPy
Jupyter
Add later or when needed
SQL
Polars
Docker
Machine learning
Classical models and the ideas behind them.
Start with
scikit-learn
Matplotlib
Add later or when needed
XGBoost
LightGBM
MLflow
Deep learning
Neural networks and pretrained models.
Start with
PyTorch
Hugging Face Transformers
Add later or when needed
TensorFlow / Keras
ONNX
CUDA basics
LLM applications
Building products on top of language models.
Start with
A hosted model API
Embeddings
Prompt design
Add later or when needed
LangChain or LlamaIndex
Agent frameworks
Model Context Protocol
Vector search
Storing and retrieving meaning, not just keywords.
Start with
pgvector or Chroma
Add later or when needed
Qdrant
Pinecone
Weaviate
Hybrid search
Deployment
Getting models and apps in front of users.
Start with
FastAPI
Streamlit or Gradio
Add later or when needed
Docker
Cloud GPU services
Monitoring tools
Priorities
What to learn first, and what can wait
Focus beats coverage. These lists keep the first months pointed at what pays off.
Learn these first
1Python until you can write and debug a 100-line script without a tutorial
2Enough pandas to clean and explore a dataset
3The core ML ideas: training versus test data, overfitting and evaluation
4How language models work conceptually - tokens, context and embeddings
5Calling a model API and getting reliable structured output
Safe to skip at the start
Advanced calculus and proofsIntuition is enough until you head toward research roles.
Training large models from scratchIt needs budgets and data most learners do not have. Start from pretrained models.
Every new framework releaseFrameworks change monthly. Concepts like retrieval and evaluation do not.
Kubernetes and MLOps platformsUseful later in engineering roles, not in your first six months.
Reinforcement learningA deep specialism. Learn it when a project or role actually needs it.
Projects
AI projects that teach the right lessons
Ordered by difficulty. Two or three finished ones beat ten half-built ones.
Beginner
Spam or not spam
Train a text classifier on a public dataset and compare it against a simple keyword rule.
scikit-learn
Evaluation
Beginner
Data story
Pick a dataset you care about, clean it and publish a notebook with clear charts and conclusions.
pandas
Visualisation
Intermediate
Image classifier with transfer learning
Fine-tune a small pretrained vision model on your own photos and report where it fails.
PyTorch
Transfer learning
Intermediate
Document Q&A assistant
Answer questions from a set of PDFs with citations to the source passages.
Embeddings
RAG
Model API
Advanced
Evaluated AI feature
Build an AI feature with a test set of real questions, automatic scoring and a report on cost and accuracy.
Evaluation
Deployment
Advanced
Tool-using agent
An assistant that can search, calculate and call an API safely, with limits on what it may do.
Agents
Tool calling
Guardrails
Avoid these
Common mistakes
Patterns that slow most learners down, and what to do instead.
Starting with the newest tool
Jumping straight to an agent framework without Python or ML basics leaves you unable to debug anything that goes wrong.
Treating maths as a gate
Many learners stall for months on textbooks. Learn the maths alongside code, when a concept needs it.
Never measuring quality
A demo that works on three examples is not a result. Build a small test set early and keep it.
Collecting certificates instead of projects
Courses help you learn, but employers check what you built. One well-explained project outweighs several certificates.
Trusting model output blindly
Language models produce confident mistakes. Verify, validate and design for the moment the model is wrong.
Trying to learn every branch at once
Vision, speech, agents and research are different careers. Pick one direction after the shared foundations.
Career
Choosing an AI career path
The field has several distinct jobs. Knowing which one you want saves months.
AI job titles overlap and change quickly, so read the responsibilities rather than the title. Roles that build features on top of models suit the AI developer path; roles that own models, data pipelines and production quality suit the AI engineer path. Strong general software skills, covered in the software development roadmap, make both far easier to land.
Typical job titles
AI Developer
AI Engineer
Machine Learning Engineer
Data Scientist
Applied AI Engineer
NLP Engineer
Job-ready checklist
You can write clean Python and use Git collaboratively
You can explain overfitting, evaluation metrics and embeddings in plain words
You have built and evaluated at least one AI feature
One project is deployed and usable by someone else
You can discuss cost, latency and failure cases of an AI system
Portfolio guidance
Three projects in your chosen direction beat ten tutorial clones
Each README states the problem, approach, results and limitations
Show evaluation numbers, even simple ones
Include a short demo video or live link
Keep API keys and private data out of public repositories
Interview preparation
Python coding and data manipulation exercises
ML fundamentals: bias-variance, metrics, data leakage
How transformers and embeddings work, at a conceptual level
Designing an AI feature: retrieval, evaluation and fallbacks
Walking through one of your projects in depth
Resources
Where to learn each part
References worth keeping open, plus the pages here that go deeper.
If you want to connect, please WhatsApp: 3265503569
One-to-one mentorship for the roadmaps on Fox Is Happy is being set up. Sessions will cover code review, project feedback, interview preparation and a learning plan matched to where you actually are - not where a generic curriculum assumes you are.
Pricing will be reasonable and listed clearly before you book - pay per session, with student-friendly rates planned. No subscriptions and no bundles you do not need.
Register interest
3265503569
WhatsApp only. Message on WhatsApp to register interest and be told first when places open.
Mention AI Roadmap in your message so the reply matches what you are working on.
Understanding the AI field before you commit
The word AI covers everything from a spreadsheet forecast to a conversational assistant. Knowing how the pieces relate makes the learning order obvious and stops you chasing whichever tool is trending this month.
The layers of artificial intelligence
Think of AI as nested circles. Machine learning is the part of AI where systems learn patterns from data instead of following hand-written rules. Deep learning is the part of machine learning that uses large neural networks. Large language models are deep learning models trained on huge amounts of text, which is why they can write, summarise and reason about language - and also why they can confidently get things wrong.
This is why an artificial intelligence learning path starts with classical machine learning even if your goal is building with language models. The vocabulary of training data, evaluation and overfitting applies everywhere, and it is what lets you judge whether a system actually works.
At 10-15 hours a week, most people with no programming background need four to six months to reach a solid foundation, and a further six to twelve months to become employable in a specialised role. If you already program professionally, you can move through the first two stages in weeks. Anyone promising job-readiness in thirty days is selling a course, not describing the field.
Do you need a degree or advanced maths?
For building AI applications, no. Many AI developers come from web or backend development and learn the AI layer on top. For research and some machine learning engineering roles, a degree in computer science, maths or a related field still helps a great deal. In every case, the maths you need first is intuition about vectors, probability and gradients - you can deepen it later.
Keeping up without burning out
New models and frameworks arrive constantly, and it is easy to feel permanently behind. The durable skills - Python, data handling, evaluation, retrieval, sensible system design - change slowly. Follow one or two reliable sources, try new tools when a project needs them, and let the rest pass. Depth in the fundamentals is what lets you pick up any new tool in an afternoon.
Where to go next
Once the shared foundations feel solid, pick your branch. Product-minded learners should continue with the roadmap for AI developers. Those drawn to models, pipelines and reliability should follow the AI engineer learning path. If your programming is still shaky, the software developer roadmap builds the general engineering habits that every AI role assumes.
Frequently asked questions
What is an AI roadmap?
It is an ordered learning plan for artificial intelligence. This one starts with Python and data, moves through machine learning, deep learning and language models, and then branches into AI development or AI engineering depending on the work you want to do.
How should a complete beginner start learning AI?
Learn Python first, then basic data handling with pandas, then classical machine learning with scikit-learn. Only after that move to neural networks and language models. Skipping straight to AI tools usually leads to confusion the first time something breaks.
What is the difference between this AI roadmap and the AI developer or AI engineer roadmaps?
This page is the broad overview and shared foundation. The AI Developer Roadmap focuses on building applications with models, while the AI Engineer Roadmap focuses on evaluating, adapting, deploying and monitoring AI systems in production.
Is machine learning still worth learning when large language models exist?
Yes. Classical machine learning is still widely used for prediction on structured data, and its concepts - evaluation, overfitting, data quality - are exactly what you need to judge whether an LLM-based system works.
Can I learn AI without a computer science degree?
Yes, especially for application-focused roles. You will need to show evidence through projects. Research roles and some specialised engineering positions still tend to favour formal qualifications.
Do I need an expensive GPU to learn AI?
No. Most learning can be done on a normal laptop, free notebook environments and hosted model APIs with small budgets. You only need serious GPU access for training or fine-tuning larger models.
