The Complete Guide to Building Python Apps with OpenAI and GPT

A project based, newcomer friendly, and all-in-one solution to learning how to use GPT3.5 & GPT 4 to solve problems, build apps, and more.

Learn OpenAI GPT & AI Development by Building GPT based Q&A App

🔨 Project Based: In this course, you will learn how to build an AI assistant powered by OpenAI's GPT technology, 🤗HuggingFace, and Streamlit. You will start with a blank app, and add features one at a time.

📜 Concepts and Theory Explained in Plain English: Before adding a new feature, you will learn just enough theory to confidently build your app. You will learn the foundational concepts of GPT and generative AI, such as Large Language Models, Prompt Engineering, Semantic Search, Finetuning, RAG, and more.

💚 AI/ML Beginner Friendly: This course is designed to be beginner friendly, so you don't need to be an experienced AI developer. You will start with a blank app, and add features one at a time. The only prerequisite is Python and Pandas knowledge.

💾 Full Project Code + Colab Notebooks Included: You will get the full project source for the projects that include real-world code samples of using OpenAI’s APIs and their best practices. You will also get all Google colab notebooks, and access to the Q&A forum if you get stuck.

How It Works

In this course, we'll take a project-based approach. We will start with a blank app, and add new features in 5 mini-projects. Before each mini-project, we will go over the theoretical concepts and mental models needed to implement new features. Each project will all build upon each other into a complete AI application. And in doing so, you'll learn what theory is applied to what situation.

🤖 Phase 1: We will start by cloning ChatGPT using a Python based web app framework called Steamlit, then integrate with OpenAI's completion API. We will learn about Streamlit's high level features.

🗺️ Phase 2: We will apply our knowledge of prompt engineering to build and evaluate an IntentClassifier to solve classification and entity extraction problems. This will allow the chatbot to differentiate and route user prompts.

📰 Phase 3: We will have our AI assistant fetch stock news articles based on semantic search The news articles could come from any source. This example will teach you how to integrate chatbots with custom datasets, and search using sentence embeddings. You will work extensively with embedding algorithms and vector databases.

💹 Phase 4: We will have our AI assistant answer stock questions such as "Why is Tesla selling off". You will learn about Retrieval-Augmented-Generation and various strategies to handle hallucinations and increase reliability of your app.

👔 Phase 5: We will have our AI assistant plot stock charts and summarize earnings transcripts. This example will show you how to let chatbots use external tools and systems.

Intended Audience

Intermediate Python developers who are interested in coding a GPT based NLP app, from ground-up, and wants to pick up the foundational concepts about working with language models.

Prior Python and Pandas experience: The ability to follow along Python is needed.

No prior ML experience needed: This course explains every concept in as plain English as possible. Some prior AI / ML exposure is helpful, but not necessary.

Only Python is used, no Javascript knowledge necessary: both the front end and the back end for our projects is going to be written in pure Python.

What You Will Learn

🤖 AI Assistant System Design Concepts

Natural language (NL) based AI Assistants need to be architected differently from traditional webapps. This course will show how to route & process NL requests.

1) End to end process of creating a non-trivial Q&A chatbot with Large Language Models

2) The foundations of GPT and generative text - Large Language Models (LLM), Prompt Engineering

3) NLU based routing & Intentclassifiers - what they are, how to build it.

💬 Prompt Engineering

Prompt engineering is the new programming. We will cover various different heursitics for writing great prompts for various tasks including classification, entity extraction, label generation, and more.

1) Prompt Engineering: different ways of crafting the perfect prompt

2) How to evaluate and choose the best prompt

3) Meta-prompting

4) Prompting to generate training labels for new tasks

5) Advanced GPT3 parameters (temperature, Top P, Top K, logit, etc)

❓ AI-based Question Answering

Using Language Models like GPT to answer questions is rife with pitfalls, such as the commonly known "hallucination" problem. We will show you various tricks to mitigate hallucination and achieve more reliable answers from Q&A bots.

1) Foundations of Retrieval Augmented Generation (RAG)

2) Foundations of Semantic Search to retrieve relevant documents pertaining to question in real time.

3) How to use word embeddings to quantify semantic similarity

4) How to use a vector database to store word embeddings

5) How to use a language model to generate answers to questions

⛑️ Reliability

In addition to handling hallucination, we will need strategies for achieving higher reliability to overcome the traditional weaknesses of language models. We go into strategies such as prompt prefixing and ensembling.

💵 Cost Optimization

Using OpenAI APIs can get very expensive. We will show various strategies for optimizing costs. Specifically, we will provide ways to think about thorny questions such as: 1) Whether to finetune GPT or not, 2) How to pick the right GPT model (cheaper model vs advanced & expensive model)

What's Included

In addition to presentations:

Code-samples & line-by-line explanations: You will get all the code samples, including Google colab notebooks

Q&A forum access: You can post questions onto the community Q&A forum to get help where you get stuck.

Full Solution Source Code: Yes, you will get the ENTIRE source code for the project so you can deploy it yourself