Complete Architecture Curriculum • 2026 Edition

Learn Artificial Intelligence

A structured, practical multi-stage roadmap designed to take you from foundational mathematics and Python to deep learning, transformers, RAG, and autonomous AI agents.

Estimated: 6–9 Months (Self-Paced) 4 Sequential Phases 10+ Capstone Projects
Phase 1 • Beginner Tier

01. AI Basics & Core Concepts

Weeks 1–3 • Foundations

Establish a rock-solid mental model of what intelligence means computationally, the history of symbolic and statistical AI, and the exact taxonomy separating AI from Machine Learning and Deep Learning.

What is AI? → Rational agents, perception, reasoning, and the formal boundaries of modern AI. ANI vs AGI vs ASI → Narrow AI vs General AI vs Superintelligence evolutionary tiers. AI vs ML vs DL → Clear architectural Venn diagrams and the mathematical evolution of models. History of AI → From the 1956 Dartmouth workshop and AI Winters to the Deep Learning revolution.
Phase 1 Milestone Project: Build a clean terminal-based expert system and formulate hypothesis-testing datasets using modern Python.

02. Mathematics & Classical Machine Learning

Weeks 4–8 • Statistical Modeling

Learn the core math that powers modern algorithms: Linear Algebra (dot products, matrices, eigenvectors), Multivariate Calculus (partial derivatives, gradients), and Probability & Statistics. Implement classical predictive models with Scikit-Learn.

Math for AI Vectors, matrices, gradient vectors, eigenvalues, Bayes theorem, normal distributions.
Python Data Stack NumPy vectorized arrays, Pandas DataFrames, data cleaning, Matplotlib/Seaborn visualization.
Supervised Algorithms Linear/Logistic Regression, Decision Trees, Random Forests, Gradient Boosted Trees (XGBoost).
Unsupervised Learning k-Means clustering, Principal Component Analysis (PCA) dimensionality reduction, t-SNE.
Evaluation & Validation Cross-validation, Bias-Variance tradeoff, Precision, Recall, F1 score, ROC-AUC curves.
Hands-on Capstone: End-to-End Housing Price or Churn Prediction pipeline with cross-validated feature pipelines deployed on Streamlit.
Phase 2 • Intermediate Tier

03. Deep Learning & Neural Architectures

Weeks 9–14 • Neural Networks

Move from statistical modeling to multi-layered artificial neural networks. Understand forward propagation, activation functions (ReLU, GELU), loss functions, backpropagation via calculus chain rule, and modern optimizers (AdamW).

PyTorch Fundamentals Tensors, Autograd automatic differentiation, nn.Module, custom DataLoaders, and GPU CUDA acceleration.
Computer Vision (CNNs) Convolutional kernels, pooling layers, ResNet skip connections, transfer learning on pretrained backbones.
Sequence Modeling (NLP) Word embeddings (Word2Vec), tokenization algorithms (BPE, WordPiece), RNNs, LSTMs, and vanishing gradients.
Regularization Techniques Dropout, Batch Normalization, Layer Normalization, weight decay, and learning rate schedulers.
Intermediate Milestone: Build and train an image classification convolutional network and a sequence-to-sequence text predictor from scratch in PyTorch.

04. Generative AI & Transformers

Weeks 15–18 • Foundation Models

Dive deep into the breakthrough 2017 Transformer architecture: self-attention mechanisms, multi-head attention, positional encodings, autoregressive generation, and how LLMs are pre-trained on internet-scale text.

What is Generative AI? → Synthesis vs discrimination, multimodal token representations, probabilistic sampling.
Self-Attention & Transformers Query, Key, Value matrix operations, scaled dot-product attention, masking, and parallel training.
Tokenization & Latent Space Byte-Pair Encoding (BPE), vocabulary dictionaries, dense semantic embeddings, and cosine similarity.
Pre-training & Alignment Next-token prediction, Masked Language Modeling (BERT), SFT (Supervised Fine-Tuning), and RLHF/DPO.
Phase 3 • Advanced Tier

05. Production Prompt Engineering & Context Design

Weeks 19–20 • Prompt Architecture

Master deterministic steering of non-deterministic models. Move beyond simple conversational prompts to structured schema outputs, few-shot conditioning, and constrained inference.

Reasoning Techniques Zero-shot CoT (Chain-of-Thought), few-shot exemplar framing, Tree-of-Thought (ToT), Self-Consistency sampling.
Structured Outputs JSON Schema mode enforcement, Pydantic validation, instructor libraries, and strict system prompts.
Security & Defenses Prompt injection sanitization, indirect injection mitigations, and model jailbreak defenses.

06. RAG (Retrieval-Augmented Generation)

Weeks 21–24 • Enterprise Grounding

Connect LLMs to private enterprise data sources to eliminate hallucinations, enforce privacy boundaries, and ground answers in up-to-date documentation.

Document Parsing & Chunking Recursive character splitting, semantic chunking, layout-aware PDF parsers, and metadata enrichment.
Embedding Models Bi-encoders (text-embedding-3, nomic-embed), dimensional truncation (MRL), and semantic similarity.
Hybrid Search & Reranking Combining dense vector search with sparse BM25 keyword search; applying cross-encoder rerankers (Cohere).
Advanced RAG Patterns Self-RAG, Corrective RAG (CRAG), hypothetical document embeddings (HyDE), and multi-query decomposition.
Advanced Milestone: Build a production-grade Financial or Legal Documentation RAG engine with hybrid search, streaming citations, and evaluation metrics using Ragas.

07. Vector Databases & Semantic Indexing

Weeks 25–26 • High-Scale Vector Search

Storage, indexing, and millisecond retrieval of high-dimensional vector embeddings across millions of data points.

Indexing Algorithms HNSW (Hierarchical Navigable Small World), IVF (Inverted File Index), and Product Quantization (PQ).
Vector Engines Cloud-native solutions (Pinecone, Qdrant, Milvus, Chroma) and PostgreSQL extensions (pgvector).
Filtering & Namespaces Pre-filtering vs post-filtering on metadata attributes, multi-tenant workspace separation.
Phase 4 • Mastery & Frontier Tier

08. Autonomous AI Agents & MCP

Weeks 27–30 • Agentic Systems

Build autonomous software entities capable of reasoning, planning multi-step actions, calling external tools, and reflecting on environment outputs to solve open-ended goals.

What are AI Agents? → Autonomous perceive-plan-act-observe loops and the 4 structural pillars of modern agents. GenAI vs AI Agents → Complete comparison matrix, architecture diagrams, and real-world execution walkthrough.
Model Context Protocol (MCP) Standardized protocol specifications for connecting foundation models to databases, local tools, and APIs.
Agent Memory Architectures Short-term context scratchpads, working buffers, and persistent episodic memory in Vector DBs.

09. AI Orchestration Frameworks

Weeks 31–32 • Multi-Agent Workflows

Work with industry-standard orchestration libraries to compose robust deterministic state machines, human-in-the-loop approvals, and specialized agent teams.

LangGraph & StateGraphs Cyclical graph execution, checkpointing, state management, and human-in-the-loop approval nodes.
LlamaIndex Workflows Event-driven workflows for ingestion, transformation, retrieval, and synthesis.
CrewAI & AutoGen Multi-agent delegation, role-playing agent swarms, and collaborative task pipelines.

10. AI Deployment, Fine-Tuning & LLMOps

Weeks 33–36 • Production Engineering

Scale, optimize, and serve models in production environments. Master model quantization, high-throughput inference engines, fine-tuning adapters (LoRA/QLoRA), and observability.

Fine-Tuning with PEFT & LoRA Low-Rank Adaptation, QLoRA 4-bit quantization, dataset curation, and Hugging Face SFTTrainer.
High-Throughput Serving vLLM PagedAttention, TensorRT-LLM, Ollama local runtimes, continuous batching, and GPU memory management.
Quantization Formats GGUF for CPU/edge inference, AWQ (Activation-aware Weight Quantization), and EXL2 for consumer GPUs.
Evals & Observability LangSmith, Phoenix Arize, OpenTelemetry tracing, hallucination rate monitoring, and cost tracking.
Mastery Capstone: Deploy a self-hosted open-weights LLM (e.g. Llama-3 or Mistral) on Cloud Run/Kubernetes via vLLM with streaming completions, rate-limiting, and LangSmith observability.

Interactive Roadmap Visual Hierarchy

Click any active branch below to jump straight into the corresponding Cyvora learning guide or topic page:

Ready to start your AI learning path?

Begin with Lesson 01: What is Artificial Intelligence? or explore all 15+ job-ready IT career roadmaps in the Learning Hub.

Start Lesson 01 → View All Hub Modules