With the rapid development of AI technologies, keeping up with the latest advancements can feel overwhelming. This video explores how these changes will impact our daily lives by encouraging you to experiment and understand AI features. To help you get started, the presenter shares seven AI project ideas ranging from beginner to advanced levels. These projects are designed to provide practical experience with AI concepts while producing tangible, real-world results.

One of the starting projects involves no coding but rather Prompt engineering with ChatGPT, allowing you to create customized GPT to serve specific purposes, like being a virtual study buddy or nutritionist. As projects advance, you can explore local LLMs for personal data analysis, build retrieval Augmented generation systems, develop engaging audio from research papers, and construct AI tools for productivity enhancement. These initiatives aim to deepen your understanding of AI's capabilities and limitations.

Takeaways from this video include:

Please remember to turn on the CC button to view the subtitles.

Key Vocabularies and Common Phrases:

1. Prompt engineering [prɑːmpt ɛndʒɪˈnɪərɪŋ] - The process of designing prompts to yield specific responses from artificial intelligence models.

The first project doesn't require any coding. The only skill you need is some Prompt engineering and a chat DPD subscription.

2. Visualization [ˌvɪʒuəlaɪˈzeɪʃən] - (n.) The representation of data in a pictorial or graphical format.

You can even create your own data Visualization expert who can create Visualization consistently in the style and the color palettes you like.

3. Augmented [ɔːɡˈmɛntɪd] - (adj.) Enhanced or increased, often referring to technology that improves human capabilities.

Develop a retrieval Augmented generation or rack system that allows users to query information from PDF documents.

4. Democratizing [dɪˈmɒkrətaɪzɪŋ] - (v.) Making something accessible to everyone; often related to equitable distribution of resources or information.

This project is exciting because you're not just building a tool, you are Democratizing knowledge.

5. Neural networks [ˈnjʊərəl ˈnɛtwɜːks] - (n.) A series of algorithms that attempt to recognize underlying relationships in a set of data through a process mimicking the way the human brain operates.

This project will help you understand what's the process of training a large language model from scratch and the basics of Neural networks.

6. Hyperparameters [ˌhaɪpərˈpærəˌmiːtərs] - (n.) Variables that determine the network structure and how the network is trained in machine learning.

You'll also learn about some important natural language processing concepts such as Tokenization and text embedding and how to fine tune important hyper parameters.

7. Tokenization [ˌtoʊkənaɪˈzeɪʃən] - (n.) The process of breaking down text into smaller units (tokens) that a language model can understand and process.

You'll also learn about some important natural language processing concepts such as Tokenization and text embedding.

8. Backward propagation [ˈbækwərd prəˈpæɡeɪʃən] - (n.) A method used in artificial Neural networks to calculate the gradient of the loss function with respect to network weights.

Building your own LM this project will help you understand what's the process of training a large language model from scratch and the basics of Neural networks like forward and Backward propagation.

9. Activation functions [ˌæk.tɪˈveɪ.ʃən ˈfʌŋk.ʃənz] - (n.) Functions used in Neural networks that decide whether a neuron should be activated or not.

Building your own LM this project will help you understand what's the process of training a large language model from scratch and the basics of Neural networks like forward and Backward propagation, Activation functions.

10. Gradient descent [ˈɡreɪ.di.ənt dɪˈsent] - (n.) An optimization algorithm used to minimize the cost function in machine learning.

You'll also learn about the Transformers architecture which is behind the LLMs nowadays, and you also learn about some important natural language processing concepts such as Tokenization and text embedding and how to fine tune important hyper parameters.

Seven AI Projects From Beginner To Advanced Ideas

Keeping up with AI developments is like playing whack a mole. OpenAI just released their strawberry model and now theyre considering $2,000 monthly subscriptions for their new LLM. So how is all this gonna change our lives?

Theres only one way to find out. Try to understand them and experiment. To me thats the best way to get through the noise and see for yourself what AI can do well and what are the limitations or challenges in todays video?

Im going to share with you seven AI project ideas from beginner to intermediate to advanced level that you can build. They are tangible and real world project to get your hands dirty while learning important AI concepts. So let's dive right in.

I present them from the easiest to the more complex projects that require more advanced skills. The first project doesn't require any coding. The only skill you need is some Prompt engineering and a chat DPD subscription.

ChatGBT allows you to create your own GPTs. You can write custom instructions for the GPT and you can also supply additional context to your GPT as custom knowledge base. Once I created my own GPT this way, uploading all my YouTube video scripts to OpenAI, they are public data anyway so but otherwise you should be careful with your personal information.

This GPT allows you to chat with me, use my knowledge, and ask anything you want related to data science using my YouTube video content. Here's the system prompt that I use to create this GBD.

With all these customizations and instructions, the GBD more likely gives you what you are looking for in a consistent way. That means you can create your own GBD as your dedicated nutritionist or trainer or a study buddy who helps you explain difficult concepts.

If you work in data analysis and data science, you can even create your own data Visualization expert who can create Visualization consistently in the style and the color palettes you like. For example, Luke Barouse has this customization for his GPT and he tells the GPT which Visualization library to use and whether to use dark theme or that it should always use a dark theme and with a certain style for the background and also for bar charts. Always order high to low and for color palettes, prioritize the certain color palettes that he likes.

If you want to do this project, you can find my tutorial for this project over here. Now if you'd rather keep your data for yourself than share it with OpenAI, theres good news for you.

You can also use an open source local LLM to create your AI project. I use a local LLM for a project where I want to analyze my bank statements and categorizing my expense items and fighting patterns in my finance in the past two years.

These open source models are free, but note that since we often can only run smaller versions of these models locally, the outputs might not be as good as some of the larger models like GPT 400 or Claude 3.5 sonnet model via their APIs.

This project will teach you how to run and use a local LLM via Olama or GPT 40. Create a model file with the desired parameters and write system prompt to control the behaviors of the LLM and you also get to practice your data dangling and data analysis skills and even create a financial dashboard for yourself.

Feel free to check out the in depth tutorial for this project on my channel link over here and to quickly write code and analyze data. These days my secret weapon is using AI coding assistance id like to shout out to my favorite free AI coding system, codium, who has kindly sponsored this video.

Codium is compatible with pretty much any ide and programming language of your choosing. I use it mostly on versus code and jetbrains ides, where it is rated five stars for both of the marketplaces. I often use codium to generate code, add documentation, and find that this really helps me stay in the flow state and remain productive.

Anyone can use codium completely for free forever, which is made possible due to the success they have had with their enterprise customers, some of which are among the biggest Fortune 50 companies in the world who trust codium for their dedication to data security and privacy.

Codium also just released a new pro tier which for the time being gives users unlimited access to some of the top language models like Cloud 3.5, Sonnet and GPT 4.0. If you want to code faster and develop your project more quickly, give Codium a try for free using my link in the description below.

Moving on to the next project. Develop a retrieval Augmented generation or rack system that allows users to query information from PDF documents.

One of the most important use cases for AI in business today is information retrieval. Extracting structured information from unstructured to your data from a custom knowledge base says like a PDF document, a report, customer invoice, and so on.

Let's admit it, most of us would want to avoid these kind of tasks in our jobs, sifting through many reports or documents, reading them, understanding them and finding the right pieces of information and organizing it, for example in an excel table.

It is tedious and very time consuming. If you have some tasks like that in your work or in your personal life. The good news is this can now be automated with great accuracy, leveraging LLMs like chat, GPD and cloud AI models.

This project will help you learn about document processing and text extraction, vector databases and text embeddings, combining retrieval with large language models using model APIs from providers like OpenAI or anthropic for generating answer and this project will teach you a lot about the challenges of building a retrieval Augmented generation system in the real world.

For example, PDF documents in the real world often have complex layouts with tables and sections that are not very easy to parse correctly. In those cases, we need to use more advanced document processing techniques to get good and reliable results.

I recently posted an in depth rack project tutorial on my channel so you can check it out here.

Alright, so far we've only talked about working with text. The next project we are going to talk about involves also audio.

One problem I noticed that also really bothers me is that research papers are often very dry and they are not accessible to laymen because of the way they are written. They are often hard to digest unless you're also in the research field yourself.

But what if we could turn that 20 page snooze fest into an engaging podcast episode? That would be really, really cool, right?

So in this project, you would try to develop a tool to convert these research papers into an engaging audio podcast, making complex ideas accessible to everyone.

This project is inspired by a tool by Google called Notebook LM, which can now turn information into engaging audio discussions. This project is exciting because you're not just building a tool, you are Democratizing knowledge. Imagine helping your grandma understand deep learning or making medical breakthroughs accessible to patients.

This project will help you learn how to use LLM models to transform text from a research paper into a podcast, as well as how to use a text to speech API, for example the one from Google, and process and display audio in your application.

Project number five create a desktop app to help you categorize and prioritize your to Dos. We've all been there, staring at a mountain of tasks, not knowing where to start.

This project is about creating your personal productivity sidekick. It helps you tackle your to do list, suggesting whats the most important and needs to be done first.

Through this project you gain experience in building desktop applications, for example using flat framework that uses Python Prompt engineering. Tell the language model to use a task management technique like the Eisenhower matrix to prioritize tasks. And then we also process the output and create an user friendly interface.

Now let's talk about something that could be a game changer for entrepreneurs and businesses alike. A market research agent this project is all about building an AI agent that can scrape customer reviews on Amazon and Google and provide accurate insights about your target market or competitors.

This project makes use of agent architecture and how it works is that instead of having an LLM generate its final output directly, an agentic workflow plumbs the LLM multiple times, sometimes using multiple LLMs together, giving them opportunities to build step by step how to provide higher quality output.

I havent had the chance to actually build this LLM agent project, still on my to do list for a while now, but ive heard some promising things about it. Andreen has shared in one of his lectures that an agentic workflow with dumber models like 3.5 significantly outperforms zero shot prompting of smart models like GPT four.

So this project will teach you how to build a custom agent that can make use of different functions or different tools and create the desired output that you want. Today, there are also many different frameworks that you can choose from to create your own agent.

And last but definitely not least, if you are feeling brave and want to build your own GPT, this project is about training a GPT model from scratch to generate songs lyrics in any style you like.

This is for people who love music like myself, but eventually you can build your GPT on any data you like. This project is inspired by this legendary video from Andre Karpathy.

It is a difficult project that requires a lot of knowledge, so I'd recommend watching this two hour video before you dive in. Building your own LM this project will help you understand what's the process of training a large language model from scratch and the basics of Neural networks like forward and Backward propagation, Activation functions, Gradient descent algorithm, and how weights are updated in the network.

You'll also learn about the Transformers architecture which is behind the LLMs nowadays, and you also learn about some important natural language processing concepts such as Tokenization and text embedding and how to fine tune important hyper parameters.

Alright, there you have it, seven project ideas thatll take you from an AI newbie to well, maybe not an expert, but definitely someone who knows that stuff. If youve tried to build anything with AI models, you probably know more than 99% of people, and I believe youll be much less anxious whenever someone tells you we are going to be replaced by AI soon.

Id love to hear which project youre most excited about and drop a comment below and let me know what you are planning to build and if you've created something cool, share it with me. I'll link all the resources in the description below, so feel free to check them out. Thank you for watching. Bye.

Artificial Intelligence, Technology, Innovation, AI Projects, OpenAI, Learning Tools