RAG, AI Agents and Generative AI with Python and OpenAI 2026

Explore python for rag and gen AI with an optional appendix python crash course for beginners, and practice functions and classes with solutions on Google Colab.

Discover how Python functions act as reusable building blocks, using parameters and returns to organize code, enable debugging, and support multi-step workflows.

Practice writing Python functions by implementing calculate_sum to sum a list and calculate_area that returns the area, in hands-on labs.

Tackle medium-level Python exercises by implementing string reversal using slicing and building an anagram finder that cleans spaces, lowercase, sorts, and compares two strings.

Build a palindrome checker by cleaning input: remove non-alphanumeric characters, convert to lowercase, and compare the string to its reverse.

Explore how classes serve as blueprints and objects as instances in Python, using a Pokemon example to show attributes like name, type, and level, plus the init method and self.

Define a book class with title, author, and pages, implement an init, and a book info method that returns an f-string; test by printing the attributes.

Define a rectangle class with length and width, implement area and perimeter methods, and test by creating an instance and printing results with f-strings.

Define a bank account class with an account holder and balance, implement deposit and withdraw with balance checks, and add a get balance method tested through simple scenarios.

Learn how retrieval augmented generation (RAG) lets large language models access your own data by retrieving relevant context before generation, powering a chatbot from your documents.

Build your first rag by combining generation and retrieval models, learn embeddings and vector stores, and assemble a no-code course assistant that accesses curriculum and course materials.

RAG combines retrieval, augmentation, and generation to keep answers up to date, reduce hallucinations, and provide sources from your knowledge base across domains like healthcare, finance, and education.

Flowise is a no-code tool to build AI racks, offering a free plan and core features like chat flows, agent flows, executions, and credentials and API key management.

Acquire an OpenAI API key by signing up or logging in, copy and securely store the secret key, and add it to FlowWise credentials to power your first chatbot.

Build your first chatbot by creating a flow wise agent with retrieval via tools, memory, and document stores, exploring chunking and an Excel-based workflow on the free plan.

Choose a course material version, either the curriculum plan or the actual curriculum. The plan lists 18 lectures, while the final includes 20 videos with quizzes and practice tests.

Learn to set up document loaders by organizing a curriculum, brand book, and scripts, then chunk Excel files per line, process chunks, and prepare embeddings via upsert.

Split large documents into sections like a table of contents, so the AI jumps to relevant chunks, boosting speed and accuracy while avoiding the lost in the middle problem.

Explore how embeddings turn words into numbers, placing related meanings near each other to power search, recommendations, chatbots, and answering questions, with context and large text training shaping the map.

Compare OpenAI embedding models, focusing on three small versus three large. Choose the three small for better pages per dollar on the free plan, with 512-dimensional embeddings.

Use vector stores to organize embeddings into a memory-like database that searches by meaning, enabling ai to retrieve relevant data and answer questions.

Choose and configure a vector store like pinecone to store embeddings using OpenAI embeddings, manage credentials, and ensure upsert and record management with a MySQL record manager to prevent duplicates.

Set up a Postgres database with Supabase, configure connection credentials and region, and manage data flow to a vector store with upsert and cleanup options for a RAG workflow.

Process Word and PDF documents using the recursive character text splitter with chunk overlap. Prepare retrieval augmented generation by upserting chunks into the Pinecone vector store.

Finish the rag workflow by adding the retriever tool, connecting vector stores and course materials, and retrieving embedded chunks to enable search and prototype sharing.

Long Rag introduces a 2024 chunking framework with massive 4000-token chunks, improving retrieval accuracy by using fewer, larger chunks and a single chunk retrieve LM step for simplicity.

Explore prompt engineering to optimize interactions with large language models by defining a system message that sets the tone, role, and goals for reliable, relevant outputs in RAG agents.

Explore how the system message acts as the rulebook that defines AI role, tone, and boundaries, shaping prompt structure, priorities, and outcomes for robust responses.

Explore the OpenAI playground chat models, including GPT five, five nano, and five mini, and learn how system and developer messages shape responses within a hidden hierarchy.

Design a system message for a RAG workflow by defining persona, goal, and restrictions, then integrate a course materials retriever to shift a chatbot from generic responses to retrieval-augmented answers.

Learn how to hide the system message and prevent prompt injection by examining developer messages and game-like challenges that reveal hidden guidelines in AI conversations.

Explore how tokenization turns words into numerical tokens and token IDs, using examples like apple and iPhone to show splits, while noting input quality shapes outputs.

Explore how temperature and top_p control how random or safe the AI's word choices are, shaping creativity and reliability. Lower values yield predictable answers; higher values boost variety.

learn how temperature and top probability relate to model parameters across GPT-5 variants and 4.1, and how to choose settings for deterministic versus creative outputs in practical applications.

Explore zero-shot, one-shot, and few-shot prompting, plus persona and goal prompts, to improve accuracy and consistency when guiding large language models.

Develop a restaurant chatbot capstone project by processing briefing materials and menu data, translating Portuguese items to English, and building an llm with a retriever tool for question and answer.

Load documents with document stores, split Excel and Word into 1000-token chunks using long rag, then upsert embeddings with Pinecone free plan and configure Supabase credentials.

Build a retrieval-augmented generation flow by wiring a retriever tool, vector store, and memory to a chat model, powering an AI waiter that queries restaurant data and handles image uploads.