PH
AI Engineer Bootcamp 2027: LLM, RAG, AI Agents, MCP (Python)
Published 11/2026
Created by Bahadır ERDUN
MP4 | Video: h264, 1920x1080 | Audio: AAC, 44.1 KHz, 2 Ch
Level: Beginner | Genre: eLearning | Language: English + subtitle | Duration: 79 Lectures ( 10h 42m ) | Size: 4.5 GB
Learn AI engineering in Python: prompt engineering, RAG, tool calling, AI agents, LangGraph, MCP, Ollama. Build Jarvis
What you'll learn
Call OpenAI-compatible APIs from Python and understand messages, roles, system prompts, tokens and cost Use OpenAI, Gemini and Ollama with one codebase, and learn how the Claude SDK differs Build a streaming chat screen with Streamlit and manage conversation history Get reliable structured output from a model with JSON and pydantic schemas Test LLM apps with a fake model and pytest, without any network calls Build semantic search and a semantic cache with embeddings and cosine similarity Build a RAG system that answers questions about PDFs: chunking, thresholds, citing sources, preventing hallucinations Build tool calling by hand and then natively, and add approval and undo to tools that write Write a think-act-observe (ReAct) agent by hand, make it safe and rebuild it with LangGraph Write an MCP server and connect your app to other MCP servers Run local models with Ollama and route requests local first, cloud when the local model doesn't know Round off the app with speech (Whisper, TTS), vision, evals, cost measurement and prompt injection defensesRequirements
Basic programming knowledge is required (variables, loops and functions in any language). The Python constructs used in the course are explained briefly the first time they appear, and there are two "Python for AI" lessons. Python 3.14 and VS Code should be installed. If you don't know how, that's fine: the first lesson shows how to install Python, VS Code and Ollama on a clean PC step by step, and a written setup guide (PDF) is included in the course files. A Windows, macOS or Linux computer. At least 8 GB of RAM is recommended for the local model lessons; a GPU is not required. An OpenAI API key (ρáíd; the calls in the lessons cost fractions of a cent, and we show how to set a spending limit). Gemini and local model (Ollama) options are covered too. No AI or machine learning knowledge is needed.Description
This course contains the use of artificial intelligence.
If you want to learn how to build a real AI product with ready-made language models, this course was made for that. An AI engineer usually does not train models from scratch; they turn ready models such as OpenAI, Gemini, Claude or a local model running on your own computer into a product with their own code, and fine-tune when needed. University "AI engineering" degrees cover the whole field; in this course we use the term in its current industry sense: the engineer who builds products on ready models. You learn it in Python, with small projects that each run on their own: prompt engineering, RAG, tool calling, AI agents (agentic AI), LangGraph and MCP; from setup to deployment in about ten hours. The course is beginner-level in AI: no AI or machine learning background is needed, we start from installing Python and the very first API call; basic programming knowledge (variables, loops, functions in any language) is enough, and two short "Python for AI" lessons cover the constructs we use.
How do we learn?
Every lesson teaches one concept and first shows the problem on screen: without a system message the model gives a long, messy answer; without conversation history it forgets your name; without RAG it makes things up; an agent without a step limit runs in circles; a poisoned PDF tricks the assistant. Then we add the few lines that fix the problem, repeat the same experiment and watch the problem go away. The code of every lesson is downloadable, and every lesson ends with a "run and watch" step: which command to type, what you will see on screen, and how to read the most common error messages.
What do we build?
The pieces you write during the lessons come together at the end asJarvis, a personal assistant: a chat screen with Streamlit, several model providers, local-first routing with the cloud as a fallback, JSON skill files, fuzzy text matching, a semantic cache, asking questions about your PDFs (RAG), user memory, weather, note and reminder tools (with approval and undo), an agent built by hand and the same agent in LangGraph, an MCP server and client, push-to-talk speech (Whisper), spoken answers (TTS), screenshot understanding and a small eval set.
Beyond Jarvis: forward deployed engineer skills. A final section takes the assistant into a customer's world with the skills of a forward deployed engineer: discovery, data integration, CI/CD, observability, demos. You turn a messy customer brief into a one-page spec and an acceptance eval, connect a SQL database, CSV/Excel and legacy feeds and a SaaS REST API, build a website assistant with FastAPI and a chat widget, and ship it with GitHub Actions, traces, API keys and a demo script.
Topics
LLM APIs: messages, roles, system prompts, tokens and cost, streaming One codebase, many providers: the OpenAI-compatible standard, Ollama, Gemini, Claude SDK differences Model settings and prompt engineering: temperature, top_p, max_tokens, seed, few-shot examples, prompt templates Structured output: getting JSON from a model, pydantic schemas Testing with a fake model (pytest): no API calls, free, done in seconds Coding with AI, and checking the tests AI writes with mutation testing Embeddings, cosine similarity, a semantic cache, picking an embedding model by measuring it RAG: chunking PDFs, similarity thresholds, teaching the model to say "it's not in the document", citing sources, and the Chroma vector database Tool calling: first by hand with a JSON protocol, then with native tool calling AI agents: the ReAct loop, step limits, pausing for approval and resuming, LangChain and LangGraph MCP (Model Context Protocol): writing your own server and connecting to other servers Local models: Ollama, hybrid routing (local first, cloud when it doesn't know), Hugging Face pipelines, LoRA fine-tuning Speech and vision: Whisper, TTS, showing the model an image Production: evals, cost and speed, prompt injection, API key security, deployment (GitHub, Streamlit Cloud, Docker), responsible AI and GDPR Forward deployed engineering: discovery and specs, system design, SQL and file-feed integration, REST APIs with retries and idempotency, a FastAPI website assistant, CI/CD with GitHub Actions, observability, access control, demos and handoffCourse language and tools: narration in English. Python 3.14, VS Code, Streamlit, the OpenAI Python SDK, Ollama, LangGraph, the MCP Python SDK. The code is shown on screen in a real VS Code view and explained line by line.
Who it's not for: this course is about building products on ready models. We train a tiny GPT from scratch and LoRA-fine-tune a model to understand what's inside, but training large models (ML engineering) and data analysis/prediction models (data science) are out of scope.
Who this course is for
Software developers who want to build products with ready-made language models Developers coming from another language (C#, Java, JavaScript.) who want to start building AI apps in Python Anyone who wants to learn RAG, AI agents, tool calling and MCP with working code People who want to build their own personal assistant or an internal AI tool for their company Developers thinking about moving into an AI engineer role Software engineers, product managers and DevOps engineers with no prior AI experience who need to build, integrate or evaluate LLM features at workYou do not have permission to view the full content of this post. Log in or register now.
You do not have permission to view the full content of this post. Log in or register now.
You do not have permission to view the full content of this post. Log in or register now.
You do not have permission to view the full content of this post. Log in or register now.
You do not have permission to view the full content of this post. Log in or register now.