Marsof Academy
SyllabusPricingSign in
Syllabus

Syllabus of the Artificial Intelligence course from scratch

No prior experience · In English or Spanish · At your own pace

The four courses and the optional advanced level, unit by unit and lesson by lesson: 181 lessons and 11 projects, from Python basics to machine learning, deep learning and LLMs.

Create my accountSee pricing

Course 1

Foundations: programming and data

From zero to programming in Python and handling real data

  • ≈ 173 h
  • Level: beginner
  • 4 units
  • 36 lessons

How a computer, the terminal and Git work; Python as an engineering tool; algorithms and data structures; and NumPy, pandas, SQL and visualisation to prepare any dataset. It is the foundation for everything else and opens the fast track to the LLM applications course. The course doesn't require English: everything is available in Spanish. Some recommended further reading only exists in English (marked "(en inglés)" in Spanish); if that's hard for you, use your browser's automatic translation.

What you'll be able to do

  • A command-line application with tests, packaged and versioned with Git
  • A data pipeline that downloads, cleans, analyses and plots real data
  • SQL queries and pandas analysis on any dataset, with no data leakage
Unit 1 · Digital foundations5 lessons
  1. How a computer works
  2. Operating system and terminal
  3. Git and version control
  4. Python, virtual environments and pip
  5. Set up your computer
Unit 2 · Engineering with Python16 lessons
  1. Variables, types and expressions
  2. Operators, formatting and precision
  3. Conditionals and loops
  4. Functions
  5. Lists and tuples
  6. Dictionaries and sets
  7. Comprehensions and the collections module
  8. Exceptions and error handling
  9. Files, CSV and JSON
  10. Classes and objects
  11. Inheritance, composition and special methods
  12. Iterators and generators
  13. Decorators and context managers
  14. Types, dataclasses and tests
  15. Advanced Python (optional): measure before you optimise
  16. Advanced Python (optional): concurrency and async
Unit 3 · Computer science8 lessons
  1. Complexity and Big O notation
  2. Arrays, linked lists, stacks and queues
  3. Hash tables and sets
  4. Recursion
  5. Sorting and searching
  6. Trees and graphs
  7. Advanced (optional): threads, races and locks
  8. Advanced (optional): distributed systems in 45 minutes
Unit 4 · Data for AI7 lessons
  1. NumPy in depth
  2. pandas (I): selecting, filtering and grouping
  3. pandas (II): joining tables, missing values and time series
  4. SQL with sqlite3 (I): querying, grouping and joining
  5. SQL with sqlite3 (II): window functions, CTEs and indexes
  6. Visualisation — seeing the data before modelling it
  7. Data cleaning and exploratory analysis

Course projects

  • CLI application: a task manager ≈ 25 h
    A command-line task manager with JSON persistence, tests and a layered design.
  • A data pipeline: Mauna Loa CO₂ ≈ 30 h
    Real NOAA data (public domain) with a header that lies: fingerprinted ingestion, schema as code, quarantine, quality checks, a feature table in SQLite, incremental idempotent loads and a DAG with retries.

Course 2

Machine learning

Just enough maths and the classic algorithms, from scratch

  • ≈ 130 h
  • Level: intermediate
  • 4 units
  • 36 lessons

A bridge to the school maths you will need (functions, powers, logarithms, sums and notation), for anyone whose maths is rusty; linear algebra, probability, calculus and optimisation implemented in code; regression, classification, trees and clustering written by hand and then with scikit-learn, the library the industry uses; and how to evaluate a model honestly, detect bias and comply with regulation.

What you'll be able to do

  • A classifier trained on real data and evaluated without cheating
  • Regression, k-NN, trees and k-means implemented from scratch, and their professional version with scikit-learn pipelines
  • A recommender and a forecasting model evaluated with time-based splits, with no leakage from the future
  • A fairness audit and a model card for a real model
Unit 1 · Maths bridge7 lessons
  1. Functions, graphs and slope
  2. Powers and roots
  3. Exponentials and logarithms
  4. Log-loss and softmax
  5. Summations, subscripts and vectors as lists
  6. Percentages, proportions and scientific notation
  7. Reading formulas without fear
Unit 2 · Maths for AI12 lessons
  1. Vectors and the dot product
  2. Matrices and matrix multiplication
  3. Derivatives: slope, limit and the power rule
  4. Differentiation rules
  5. Gradient descent
  6. Probability and conditional probability
  7. Bayes' theorem
  8. Statistics and distributions
  9. Partial derivatives and the gradient
  10. The chain rule
  11. SGD, momentum and Adam
  12. Linear transformations and eigenvalues
Unit 3 · Machine Learning13 lessons
  1. Linear regression with several variables
  2. Logistic regression and classification
  3. Model evaluation
  4. Overfitting, underfitting and regularisation
  5. k-NN and decision trees
  6. Unsupervised learning
  7. Gradient boosting and XGBoost
  8. Support vector machines (SVM)
  9. Anomaly detection
  10. Recommender systems
  11. Time series and forecasting
  12. The professional version: scikit-learn
  13. Choosing a model without cheating: cross-validation and GridSearchCV
Unit 4 · AI ethics, bias and regulation4 lessons
  1. Bias and fairness in ML models
  2. Privacy and data protection in AI
  3. The EU Artificial Intelligence Act (AI Act)
  4. Responsible deployment — document, oversee, attack and respond

Course projects

  • An ML model with real data: Palmer penguins ≈ 38 h
    Real, messy field data (CC0): cleaning, a leak-free split, a baseline, logistic regression with NumPy, cross-validation, calibration, a model card and the professional version with scikit-learn.

Course 3

Deep learning and Transformers

From the neuron to your own language model

  • ≈ 287 h
  • Level: intermediate to advanced
  • 6 units
  • 36 lessons

Neural networks with backpropagation written by hand, professional PyTorch, computer vision, natural language processing and the Transformer piece by piece, up to training and serving a small LLM.

What you'll be able to do

  • A neural network from scratch, with no deep learning libraries
  • A CNN that recognises digits and an NLP system on real text
  • A complete Transformer and TuLLM, a language model you train yourself
Unit 1 · Deep Learning9 lessons
  1. The artificial neuron
  2. Multilayer networks (MLP)
  3. Computational graphs and autograd
  4. Backpropagation
  5. Training a network
  6. Activations, initialisation and normalisation
  7. Dropout and regularisation in networks
  8. Recurrent networks and the gradient through time
  9. LSTM and GRU — memory with gates
Unit 2 · Engineering with PyTorch6 lessons
  1. Tensors and broadcasting
  2. Autograd over tensors
  3. Modules, parameters and optimisers
  4. Datasets and DataLoaders
  5. A professional training loop
  6. GPU, mixed precision and checkpoints
Unit 3 · Computer vision4 lessons
  1. Convolution — a magnifying glass that sweeps across the image
  2. Convolutional networks — stacking ever wider views
  3. Detection and segmentation — what's there and where it is
  4. Vision Transformers — an image is worth 16 × 16 words
Unit 4 · Language processing5 lessons
  1. Text as data
  2. BPE tokenisation from scratch
  3. N-gram language models
  4. Embeddings — tokens as vectors
  5. Attention — every token looks at the others
Unit 5 · Transformers5 lessons
  1. Embeddings and positional encoding — from ids to vectors that know where they are
  2. Multi-head attention — many views at once, without a single loop
  3. The Transformer block — residual, LayerNorm and MLP
  4. The full decoder — from ids to logits, loss and generated text
  5. Training a small language model — data, gradients and curves that talk
Unit 6 · LLM engineering7 lessons
  1. Pretraining and scaling laws
  2. Decoding — from logits to text
  3. Inference and the KV cache
  4. From base model to assistant — SFT and DPO
  5. LoRA and QLoRA — fine-tuning without touching the weights
  6. Quantization — fewer bits, almost the same model
  7. Evaluating LLMs without fooling yourself

Course projects

  • A neural network from scratch ≈ 35 h
    A multilayer neural network with hand-written backpropagation, trained on real data.
  • A Transformer from scratch ≈ 50 h
    Implement and train a complete Transformer, verifying each component against a reference.
  • TuLLM — a language model from scratch ≈ 80 h
    Your long-haul project. Build a complete language model, from raw text to optimized inference, as you progress through the academy.

Course 4

Applications with LLMs

APIs, RAG, agents and LLMs in production

  • ≈ 224 h
  • Level: intermediate
  • 4 units
  • 30 lessons

The daily work of an application AI engineer. Serve models as reliable APIs, search your documents with RAG, build agents with tools and take it all to production with evals, guardrails, cost and observability. You can start right after course 1; whatever you need about how models work inside is explained "just enough to get started".

What you'll be able to do

  • An AI API with streaming, rate limits and Docker deployment
  • A RAG system over your documents, with hybrid search and evaluation
  • A secure agent with tools and a production support assistant with evals and traces
Unit 1 · APIs and services for AI5 lessons
  1. HTTP and REST — the language of services
  2. Building an API with FastAPI
  3. Serving a model or an LLM behind an API
  4. Deploying — Docker, configuration and health checks
  5. Your first real call to an LLM
Unit 2 · RAG7 lessons
  1. Ingestion and chunking — getting documents ready to be found
  2. Lexical retrieval — inverted index, TF-IDF and BM25
  3. Dense retrieval — embeddings, top-k and approximate indexes
  4. Hybrid search, reranking and query rewriting
  5. Building and evaluating a RAG system — prompts with citations, metrics and failure analysis
  6. Graph RAG — retrieving by following relationships
  7. Multimodal RAG — tables, images and PDF layout
Unit 3 · AI agents6 lessons
  1. Tool calling and structured outputs
  2. The agent loop
  3. Memory and context management
  4. Workflows, orchestration and MCP
  5. Agent evaluation and security
  6. Multi-agent systems — coordinating without losing control
Unit 4 · Engineering LLM applications in production12 lessons
  1. Serious prompt engineering — templates, few-shot and versions
  2. Structured outputs — JSON you can trust
  3. Function calling in depth — tools that hold up in production
  4. Evals for LLM applications — knowing whether your change improves anything
  5. Guardrails and security — layered defence
  6. Cost and latency — the bill and the clock
  7. LLMOps observability — seeing what happens inside
  8. RAG in production — freshness, permissions and citations that hold up
  9. Agents in production — budgets, humans and resumption
  10. Prompting, RAG or fine-tuning? Deciding with judgment
  11. Multimodal and voice in applications
  12. Capstone project — an end-to-end support assistant

Course projects

  • A REST API for a model ≈ 30 h
    A real FastAPI service around a model: Pydantic v2 validation, API keys, uniform errors, cursor pagination, rate limiting, OpenAPI, tests with TestClient, health checks and a Dockerfile.
  • A RAG system ≈ 45 h
    A retrieval-augmented system with indexing, hybrid search, citations and evaluation.
  • An AI agent ≈ 45 h
    An agent that plans, uses tools and recovers from errors, with an evaluation of its reliability.

Course 5

Advanced level (optional): specialisations

GPUs, performance, scale, MLOps, research and the frontier

  • ≈ 360 h
  • Level: advanced
  • 8 units
  • 43 lessons

Optional, to go beyond the "from zero to AI engineer" path (courses 1–4). GPUs and CUDA, LLM performance, distributed training, MLOps, research engineering, multimodal AI, the frontier of the field and a final AI systems project. Each unit is an independent specialisation with its own certificate; pick the ones that fit your path, you don't need to do them all. The GPU, distributed training and paper reproduction projects are marked "Coming soon".

What you'll be able to do

  • An inference server with batching, KV cache and quantisation
  • A complete AI platform, from data to deployment and monitoring, assessed with a rubric
  • Coming soon: guided projects on GPU inference, distributed training and reproducing a paper
Unit 1 · GPUs and CUDA4 lessons
  1. GPU architecture
  2. The CUDA programming model
  3. Roofline and kernel optimisation
  4. The memory hierarchy in depth — caches, loop order and tile shape
Unit 2 · LLM performance5 lessons
  1. Training and inference memory
  2. Quantization, pruning and distillation
  3. Batching, throughput and latency
  4. Tensor and pipeline parallelism
  5. Kernel fusion and FlashAttention — the online softmax
Unit 3 · Distributed AI7 lessons
  1. Data parallelism and all-reduce
  2. Model and optimizer sharding (ZeRO and FSDP)
  3. Fault tolerance, checkpoints and resumption
  4. Tensor parallelism in depth — the complete Transformer block
  5. Pipeline schedules — GPipe, 1F1B and interleaved
  6. Mixed precision, loss scaling and overlapping communication
  7. Distributed inference — serving a model that doesn't fit on one GPU
Unit 4 · MLOps and infrastructure7 lessons
  1. Serving models in production
  2. Observability, percentiles and drift
  3. Reproducibility, tracking and CI/CD for models
  4. Cost, caching, secrets and privacy
  5. Paged KV cache — virtual memory for serving LLMs
  6. Quantizing for serving — what speeds up, what fits and what breaks
  7. Progressive delivery — shadow, canary and rollback
Unit 5 · Research Engineering5 lessons
  1. Reading papers like a researcher
  2. Experimental design, seeds and uncertainty
  3. Ablations, benchmark hygiene and reproduction
  4. Technical writing, honest figures and uncertainty
  5. Reproducing a paper — from the PDF to a number you can defend
Unit 6 · Multimodal AI5 lessons
  1. Contrastive learning and CLIP — images and texts on the same map
  2. Audio and speech — from the microphone to the mel spectrogram
  3. Vision-language models and multimodal agents
  4. Generative models — autoencoders, VAEs and GANs
  5. Image generation — diffusion models
Unit 7 · The AI frontier8 lessons
  1. Reasoning and inference-time compute
  2. Long context and efficient attention
  3. Mixture of Experts
  4. Alignment and synthetic data
  5. The reward model — from human rankings to a number
  6. RLHF with PPO — reward with KL, advantages and clipping
  7. DPO — the full derivation and why it works
  8. Evaluating alignment — reward hacking, over-optimization and biased judges
Unit 8 · Final project — AI systems engineering2 lessons
  1. Designing an AI system end to end
  2. Technical defence — justifying every decision with numbers

Course projects

  • An inference server ≈ 60 h
    A server that serves a model with batching, streaming, metrics and load limits.
  • A complete AI platform ≈ 150 h
    A production AI platform that brings together data, retrieval, agents, serving, observability, evaluation and deployment, assessed with a public rubric, your self-assessment and the AI tutor's review of your technical defence.

Estimated hours for someone starting from scratch: lessons with practice and review, challenges, exams and the recommended projects. It's the same estimate you'll see inside the academy. In total, about 1174 hours of study.

Shall we start?

Create your account and activate your subscription: you get the whole syllabus from day one.

Create my accountSee pricing