AI Engineering

Course Objective

To teach participants how to design, develop, test, and implement modern AI solutions based on Large Language Models (LLMs).

The course focuses not on using ready-made AI tools, but on an engineering approach to creating AI systems: from working with APIs and prompt engineering to RAG systems, AI agents, tool calling, evaluation, security, and deployment to production.

Who is this course for?

Software developers and Full-Stack Developers who want to transition into AI
Engineering professionals
Python developers
Backend developers
Data Engineers and Data Scientists
QA Engineers who want to work with testing AI systems
Solutions and software architects
Technical product managers and technical leads
IT professionals who are already working with LLMs and want to move from experiments to creating production-ready AI solutions

Prerequisite Knowledge

Programming experience, understanding of APIs, Git, and working with JSON.

Course Outcomes

Upon completion of the course, participants will be able to:

understand the architecture of modern LLM-based systems;
work with LLMs via API and SDK;
create reliable prompts and structured outputs;
build AI applications using Python;
connect AI to corporate data;
create RAG systems and vector search;
add logging, tracing, and observability;
evaluate the latency and cost of AI solutions;
deploy AI applications to production;
design AI solutions considering security, scalability, and reliability.

What will you get from the course?

🔗 View benefits

Program

Basics of AI engineering and large language models

What is AI engineering
The role of AI engineer / AI engineer and differences from ML engineer / ML engineer, Data Scientist and Software Engineer
Architecture of modern AI systems
Basics of generative AI
Large Language Models (LLM)
How text generation works
Tokens and context window
Model parameters and their influence on the result
Choosing a model for a specific task
Commercial and open models

Development of AI applications and working with APIs

Interaction with LLM via API
API keys and secure configuration
Structure of requests and responses
System, user and development instructions
Structured answers
JSON format
Call functions
Streaming responses
Error handling
Repeat requests and backup scripts
Work with several models

Query design and context management

Basics of effective query design
Zero-shot and few-shot approaches
Role and task instructions
Controlling the format and structure of the response
Projecting the context
Managing a large amount of context
Request templates
Query versioning
Typical causes of unstable responses
Optimization of requests for quality and cost

Vector representations and semantic search

What are embeddings?
Semantic similarity
Vector databases
Splitting documents into fragments
Data fragmentation strategies
Metadata
Search by similarity
Key and semantic search
Hybrid search
Filter results
Reordering results
Practical construction of semantic search

RAG systems and work with corporate data

What is RAG?
Why is RAG needed?
The architecture of the RAG system
Data loading and preparation
Processing of documents
Fragmentation of documents
Creation of embeddings
Indexing
Finding relevant information
Formation of the context
Generating a response
Links to sources
Advanced RAG systems
Multi-step search
Typical RAG errors
Optimizing search and response quality

AI agents and tool usage

What is an AI agent?
AI Application, Workflow and Agent: Differences
Architecture of an AI agent
Agent work cycle
Planning and execution of tasks
Call tools
Call functions
Connecting external APIs
Work with databases
Connection of external services
Agent memory
Context control
Agent autonomy control
Human-agent interaction
Error handling and termination conditions

MCP and integration of AI with external systems

What is the Model Context Protocol
MCP architecture
MCP servers and clients
Tools, resources and templates
Connecting AI to external systems
Using MCP in AI agents
Integration with corporate tools
Security of MCP integrations
Practical construction of integration

Multi-agent AI systems

What is a multi-agent system
Single-agent and multi-agent architectures
Distribution of roles between agents
The architecture of "manager - executors"
Planner and executors
Delegation of tasks
Interaction between agents
Shared context
Exchange of information between agents
System complexity control
Typical problems of multi-agent solutions

Testing and evaluation of AI systems

Why AI systems can't be tested in the same way as traditional software
AI evaluation strategy
Reference datasets
Test scenarios for LLM
Evaluation criteria
Automated assessment
Assessment by LLM
Human assessment
Detection of hallucinations
Evaluation of compliance with sources
RAG testing
Testing AI agents
Regression testing
Quality metrics
Building an automatic assessment process

Security and protection of AI systems

Attacks through requests
Data leak
Protection of personal and confidential data
Bypassing model limitations
Verification of input and output data
Protective mechanisms
Access control of AI agents
Safe tool calling
A person in the decision-making circuit
Abuse of AI systems
Responsible use of AI
Safety at the design stage

Monitoring and operation of AI systems

What is LLMOps?
Journaling
Execution tracing
Monitoring
Monitoring the work of AI agents
Use of tokens
Response time
Control of the cost of requests
Diagnostics of AI applications
Manage model versions
Query versioning
Quality assessment in production
Detection of quality deterioration

Deployment of AI solutions in production

The architecture of production solutions
Creation of the server part of the AI application
FastAPI and AI services
Docker
Environment configuration
Managing secrets
Deployment of AI applications
Scaling
Caching
Limiting the frequency of requests
Backup scenarios
Cost optimization
Reliability of AI systems
Cloud deployment basics

Practical project: from idea to production

13-16

Formulation of the task and requirements for an AI solution
Choice of architecture
Model selection
Work with data
Implementation of RAG
Integration of tools
Creating an AI agent
Adding protective mechanisms
Testing and evaluation
Monitoring
Optimizing performance and cost
Deployment of the solution
Presentation of the finished AI product

FAQ

Do you need to be able to program?

Yes, programming skills are important. The course is aimed at technical specialists, so a large part of the practice involves working with programming, APIs and code.

Do you need to know Machine Learning?

No. The course is focused on AI Engineering and building applications based on ready-made models, rather than learning your own LLM from scratch.

Shall we create AI agents?

Yes. AI Agents is one part of the program: tool calling, memory, workflows, MCP, multi-agent architectures and human-in-the-loop.

Will there be a RAG?

Yes. Participants will build RAG solutions from ingestion and embeddings to retrieval, reranking, generation and evaluation.

Shall we test AI systems?

Yes. A separate module is dedicated to evaluation, regression testing, hallucination detection, LLM-as-a-Judge and evaluation of AI agents.

Is the course suitable for corporate training?

Yes. The program can be adapted to the specific technological stack and business tasks of the company — from internal AI Assistant to corporate RAG and agentic systems.

What will the participant have after the course?

The main result is not a set of theoretical knowledge, but a working end-to-end AI project that demonstrates AI Engineering skills from architecture to deployment.