Image by Author
Most people who lack technical knowledge may think that working with AI or LLMs (large language models) is challenging and reserved for experts and engineers. However, what if I told you that all you need is proficiency in Python to build a wide range of LLM projects, from Q&A systems to YouTube summarizers? You can even create your own GPT-4o application using multiple open-source models and components.
In this project, we will explore interesting and easily achievable LLM project ideas that you can build using free or affordable resources. Furthermore, each project idea is accompanied by a sample project link that you can examine to better understand how it works.
1. Fine-Tuning Llama 3 and Using It Locally
The Fine-Tuning Llama 3 and Using It Locally is a proper project with multiple steps and files. The goal is to fine-tune the model on a dataset of patient-doctor conversations using free resources provided by Kaggle. Once the model is successfully fine-tuned, it can answer medical-related questions in a highly professional manner.
Image from the project
In order to use the model offline on your laptop, you can follow these steps:
- Merge the adopter layer to the base model.
- Convert the model into the Llama.cpp format, known as GGUF.
- Reduce the size of the model using the quantization method.
- Finally, use the model on your laptop using the Windows Jan application.
It’s important to keep medical conversations private between doctors and patients, which is why it’s necessary to use it locally and ensure privacy.
2. Q&A Retrieval System
If you prefer not to fine-tune the model, you can still create context-aware AI applications locally using tools like LangChain, Chroma DB, and Ollama. This application will utilize your dataset as context before generating the response.
To build the RAG (Retrieval-augmented generation) application, you can follow these steps:
- Load PDF Files: Begin by loading all the PDF files from the designated folder.
- Split Text: Split the text into smaller chunks for efficient processing.
- Convert to Embeddings: Convert the text into embeddings and store them in the vector database.
- Build Retrieval Chain: Construct a retrieval chain using LangChain.
- Develop Python Application: Create a proper Python application to ensure a seamless chat experience.
Image from project
LangChain simplifies the process by providing a high-level API and easy-to-use commands. By following the tutorial on “How to Run Llama 3 Locally,” you can build a context-aware smart LLM application.
3. Serving an LLM application as an API endpoint using FastAPI in Python
In this project, you will be building an English to French translator API using the OpenAI API and FastAPI. The project will be divided into two main parts: understanding how to use the OpenAI API to ensure the generated output is always in French, and building a REST API using FastAPI to take in text and generate output through a simple CURL command.
If you are familiar with FastAPI, you can build an even better LLM application that can serve as an API endpoint in 30 minutes.
Image from project
For reference, follow the guide “Serving an LLM application as an API endpoint using FastAPI in Python” and try serving your LLM. It can also be a model that you are running locally.
4. Vacation Planning Assistant
Planning a vacation without a travel agent can be difficult. There are so many moving parts, and sometimes people don’t even know what to do. So, why not create your own travel itinerary application that will show your itinerary on a map and provide detailed plans and various attractions?
In this project, you will build a web application that takes user instructions about their travel plans and provides itinerary suggestions, showing them on a map.
Image from project
The project requires you to learn a few basics about Gradio, Launching, and the Google Maps API before diving into the project. You can start by following the “Building a Smart Travel Itinerary Suggester with LangChain, Google Maps API, and Gradio (Part 1)” tutorial, but if you want to build the Vacation Planning Assistant, you might have to add more components to your application and make it more robust.
5. YouTube Summarizer
YouTube Summarizer is a beginner-friendly project, perfect for students and newcomers to APIs and natural language processing. The project involves using the YouTube API to extract transcripts from videos and the OpenAI API to summarize these transcripts. Given that some videos can be lengthy, and the context window of models like ChatGPT might be limited, the project requires splitting the transcript into manageable parts. Each part is summarized individually, and the summarized sections are then combined to produce a coherent summary of the entire video.
Image from project
You can follow the “Create Your Own YouTube Video Summarizer App in Just 3 Easy Steps” tutorial and experience the awesomeness yourself.
Note: The project is using the old API. You can always check the OpenAI API documentation to update to the new structure.
6. Web scraping with LLMs
Web scraping can be a lucrative business, with individuals earning up to $200 per day by running a simple script. It is considered lucrative because it can be challenging to bypass certain website structures. In such cases, building an LLM-powered web scraper using Scrapy and Ollama can help automate or enhance web parsing.
Image from books.toscrape.com
By following the guide “Web Scraping with LLMs,” you can learn to use LLM on each webpage to extract specific attributes such as product name and price. The LLM eliminates the need for manual coding to extract these attributes from the webpage; all you need to change is the prompt.e prompt.
7. Build GPT4o using Open Models
Building an all-in-one AI application typically requires millions of dollars and years of research. What if I told you that you can build your own GPT-4o model using an open-source model at no cost and in just one day?
In this project, we will create a comprehensive Open GPT-4o application that can understand audio, image, and text data. It will include a live voice chat feature and video chat capabilities. Additionally, you can use it to generate images and videos. In short, it will be your AGI (Artificial General Intelligence) application.
Please note that the project does not come with a guide or tutorial, so you will have to learn everything by understanding the source code: app.py · KingNish/OpenGPT-4o at main (huggingface.co)
Image from OpenGPT 4o
Before you build your LLM application, I highly recommend that you test out OpenGPT 4o. Learn about various features and what type of model it is using. Learn how efficient and fast it is.
Final Thoughts
Building an LLM portfolio project can significantly boost your career prospects. If you are a student seeking employment, these 7 projects will help you secure a job faster than others. Recruiters and HR managers are particularly impressed by projects that incorporate the latest technologies, such as AI.
To begin, bookmark this page and start building the simple projects. As you progress to more complex projects, make sure to consistently showcase your work on LinkedIn. This way, you will soon catch the eye of recruiters.
Abid Ali Awan (@1abidaliawan) is a certified data scientist professional who loves building machine learning models. Currently, he is focusing on content creation and writing technical blogs on machine learning and data science technologies. Abid holds a Master’s degree in technology management and a bachelor’s degree in telecommunication engineering. His vision is to build an AI product using a graph neural network for students struggling with mental illness.
