Photo by Author | Canva
. Introduction
This is the second article in my initial project series. If you haven’t seen the first in the midwer, it is worth checking: 5 Recreation Projects for Absolute Early Persons.
So, what is it? Generative A. Or General Ai? It’s all about making new content such as text, photos, codes, audio, or even video using AI. Before the era of large language and vision models, things were quite different. But now, with the rise of foundation models like GPT, Lalma and Lalawa, everything has changed. You can create creative tools and interactive apps without training models from the beginning.
I have chicked them 5 projects to cover up a little bit of everything: Text, icon, sound, vision, and some backward concepts such as fine toning and rags. You will need to try both API -based solutions and local setups, and by the end, you must have touched all building blocks used in most modern general AI apps. So, let’s start.
. 1. Directed Generator app (Text Generation)
Link: Create a prescription generator with reaction and AI: The code matches the kitchen
We will start with some easy and entertainment that only uses text generation and an API key, does not require heavy setups. This app allows you to input some basic details such as ingredients, eating type, preference, cooking time and complexity. It then produces a complete prescription using GPT. You will learn how to create a front and form, send the data to GPT, and the AI will submit the infield directive to the user. Another modern version of the same idea is: Make AI prescription searching with GPT O1-PRIVIVIEW in 1 hour. It has more modern quick engineering, GPT -4, suggestions, alternatives to components, and more dynamic front ends.
. 2. Image generator app (stable dispersion, local setup)
Link: Make AII AIIMAGE Generator in 15 minutes (free and local)
Yes, you can just create cool images using tools such as Chattagpat, Put · E, or Midjorney by typing just one hint. But what if you want to take it a step forward and run everything locally with no API cost or cloud restrictions? This project does exactly what. In this video, you will learn how to set up a stabilizer on your computer. The creator keeps it extremely easy: you install the tiger, clone the lightweight web UI Ripo, download the model checkpoint, and run the local server. Just After that, you can enter the text indicators in your browser and create AI images immediately without the Internet or API calls.
. 3. Medical chat boot with voice + vision + text
Link: Create AI Voice Assistant App using Multi -Moodle LLM Llava and Swear
This project is not specifically made as a medical chat boot, but the use of the use fits well. You talk to him, it listens, it can see a picture (such as X -ray or document), and it responds intelligently by all three ways: sound, vision, and text. It is made in a Grade interface using Lilova (a multi -modal vision language model) and whispers (text model from an open speech). The video runs by setting it on Kolab, installing libraries, making lalawa quantity to run on your GPU, and sewing it with GTTS for audio answers.
. 4. Fine toning modern llm
Link: Excellent lyrics JEMA 3, Kevin 3, Lalama 4, PHI 4 and Incorrect with Transformers and Transformers
So far, we are using off -shelf models with quick engineering. It works, but if you want more control, the fine toning is the next step. This video of the Trailes Research is one of the best places there. So, instead of suggesting a project that just changes an excellent model, I wanted to focus on the actual process of fixing the model yourself. In this video you have shown how to improve misunderstandings using Jima 3, Kevin 3, Lalama 4, PHI4, and wrong small (library for fast, memory -efficient training) and transformer. It’s tall (about 1.5 hours), but also worth it. You will learn when fine toning means, how to prepare datases, run quick eules using VLM, and debug on real training issues.
. 5. Make local rags from the beginning
Link: Local recovery increased generation (RAG) from the beginning (step -by -step lesson)
Everyone likes a good chat boot, but most of them are separated when they are asked about things beyond their training data. The vein is useful at this place. You give your LLM a vector database of relevant documents, and it pulls the context before responding. The video runs by creating a fully local rig system using a collib notebook or your own machine. You will load the documents (such as textbook PDF), divide them into pieces, prepare embedded with phrases transformer models, save it in SQLITE-VSS, and connect them all to the local LLM (such as Lama 2 Olama). I have seen this initially I have seen initially, and once you do this, you will understand that Chat GPT plugin, AI search tools, and internal company chat boats really work.
. Wrap
Each of these projects teaches you something necessary:
Text → Image → Sound → Fine toning
If you are just going to General AI and really want to make things, not just play with the demo, then this is your blueprint. Start with what makes you most excited. And remember, it’s okay to break things. That’s how you learn.
Kanwal seals Kanwal is a machine learning engineer and a technical writer who has a deep passion for data science and has AI intersection with medicine. He authored EBook with “Maximum Production Capacity with Chat GPT”. As a Google Generation Scholar 2022 for the APAC, the Champions Diversity and the Educational Virtue. He is also recognized as a tech scholar, Mitacs Global Research Scholar, and Harvard Vacod Scholar as a Taradata diversity. Kanwal is a passionate lawyer for change, who has laid the foundation of a Fame Code to empower women in stem fields.