Skip to content
Miraat·dweb developer journey, reflected
IT EN FR عربي
AI

AI Engineering

Build with large language models: prompts, RAG, function calling, evaluation, deploy.

18 skills 6 months Advanced
This is not a theoretical machine-learning course or a math course. It is the new craft: integrating AI models like Claude or GPT into real products, managing prompts, context, evaluation, cost, and fallback.

You'll cover practical prompt engineering, embeddings and retrieval-augmented generation (RAG), tool use / function calling, multi-step agents, response evaluation, and the safety patterns that keep the bill and data leaks under control.

Outcomes: AI engineer, AI product engineer, consultant for AI integrations.
How LLMs Actually Work · 25 min FOUNDATION · 25 MIN How LLMs Actually Work Prompt Engineering Basics · 30 min FOUNDATION · 30 MIN Prompt Engineering Basics Designing System Prompts · 25 min CORE · 25 MIN Designing System Prompts Few-Shot Prompting · 20 min CORE · 20 MIN Few-Shot Prompting Structured Outputs (JSON) · 25 min CORE · 25 MIN Structured Outputs (JSON) Tool Use / Function Calling · 35 min ADVANCED · 35 MIN Tool Use / Function Calling The Claude API · 30 min CORE · 30 MIN The Claude API The OpenAI API · 25 min CORE · 25 MIN The OpenAI API Prompt Caching · 20 min ADVANCED · 20 MIN Prompt Caching Streaming Responses · 20 min CORE · 20 MIN Streaming Responses Embeddings Fundamentals · 25 min CORE · 25 MIN Embeddings Fundamentals Vector Databases · 30 min CORE · 30 MIN Vector Databases RAG Basics · 35 min ADVANCED · 35 MIN RAG Basics Advanced RAG Retrieval · 30 min ADVANCED · 30 MIN Advanced RAG Retrieval Multi-Step Agents · 35 min ADVANCED · 35 MIN Multi-Step Agents Evaluating LLM Outputs · 30 min ADVANCED · 30 MIN Evaluating LLM Outputs Cost Optimization · 20 min CORE · 20 MIN Cost Optimization Safety and Guardrails · 25 min ADVANCED · 25 MIN Safety and Guardrails

Foundation

1
How LLMs Actually Work
Tokens, context windows, sampling — the mental model that makes everything else click.
25 minutes
2
Prompt Engineering Basics
Clear instruction, role, context, examples. The boring techniques that work.
30 minutes

Core

3
Designing System Prompts
Persona, scope, refusal patterns, output format. Building the model's personality.
25 minutes
4
Few-Shot Prompting
Showing examples instead of describing — when 3 demos beat 30 lines of instructions.
20 minutes
5
Structured Outputs (JSON)
Forcing valid JSON with schemas. Tool/function calling, JSON mode, validation.
25 minutes
7
The Claude API
Anthropic SDK, messages API, system, max_tokens, stop_sequences. The essentials.
30 minutes
8
The OpenAI API
Chat Completions API, Responses API, tools, comparison points with Claude.
25 minutes
10
Streaming Responses
SSE, partial-text rendering, perceived speed wins. Implementing it cleanly.
20 minutes
11
Embeddings Fundamentals
Turning text into vectors, cosine similarity, picking models, dimensionality.
25 minutes
12
Vector Databases
pgvector, Qdrant, Pinecone. ANN indexes, metadata filtering, hybrid search.
30 minutes
17
Cost Optimization
Smaller models, prompt caching, tighter outputs, batch APIs. Where the bill goes.
20 minutes

Advanced

6
Tool Use / Function Calling
Letting the model call your code: tool schemas, multi-turn flow, error handling.
35 minutes
9
Prompt Caching
Reusing the static part of your prompts to cut latency and cost dramatically.
20 minutes
13
RAG Basics
Chunk, embed, retrieve, ground. The architecture that makes LLMs answer your data.
35 minutes
14
Advanced RAG Retrieval
Reranking, query rewriting, hybrid retrieval, when to add a small tuned reranker.
30 minutes
15
Multi-Step Agents
Loops, planning, reflection, tool selection. Pitfalls of "let the model figure it out".
35 minutes
16
Evaluating LLM Outputs
Datasets, golden answers, LLM-as-judge, regression suites. Treating prompts like code.
30 minutes
18
Safety and Guardrails
Input filtering, output checks, jailbreak resistance, abuse detection.
25 minutes