Illustration by Author | Source: Flaticon

To enter the data science job market, it’s a mistake to take for granted that a degree is enough to obtain a job. One of the main suggestions is to build a strong portfolio with personal projects that can play an essential role in standing out from the crowd and impressing the recruiter. 

With the advent of generative AI tools, like ChatGPT, a collection of standard projects, like object detection and recommendation systems, are not enough anymore to capture the attention of the company. In the last months, companies are opening positions for people able to build Generative AI solutions.

For these reasons, we are going to explore 7 project ideas that use Large Language Models for solving the task:

1. Create a Portfolio Website 
2. Personalized Voice Assistant 
3. Build your own AI translator
4. Analyze research papers
5. Creating Code documentation
6. Automate Powerpoint presentations
7. Sentiment Analysis of Reviews

1. Create a Portfolio Website 

There are a lot of tutorials that explain how to build a data science portfolio website, but it can be really intimidating for getting started from scratch without any knowledge of HTML and CSS. I tried personally, and it gives a lot of satisfaction when you reach the goal, but it took me a week between finding the right resources and putting into practice what I have learned. 

Now, with the boom of Large Language Models, you don’t need to make an effort anymore. You just need a good idea, ask questions to ChatGPT, which will return the code for your website.You can just begin with a prompt like this:

I decided to build a static website. Can you generate HTML code for building the website? Moreover, I need to have three pages: a page with my name and a short presentation, a page with my data science projects and a page with my work experience. In addition to these pages, I want a vertical navigation menu at the left to move from a page to the other.

Like in other applications, you need to have clear ideas of what you want to generate your Portfolio Website.

Project link: Build a Data Science Portfolio Website With ChatGPT

2. Personalized Voice Assistant 

In my personal life, I use Google Assistant to ask to reproduce music of different genres. For example, “Google, I want to listen to rock music” and it instantly reproduces a random song from youtube music. It’s really more rapid than writing the title of the song and it learns your preferences the more it collects your data. Wouldn't it be cool to do it as a personal project? This project can be done easily by using GPT-3 to answer the question and Whisper API to transcribe the audio.

Project link: Personalized Voice Assistant with GPT and Whisper

3. Build your own AI translator

Are you tired of copying and pasting the text into Google Translate? Personally, I have also tried google chrome extensions to translate the text on web pages, but I still struggle when I have to read PDF files in English. A possible alternative is to build your own AI application. Every day, there is a new powerful Large Language Model that amazes us with its incredible results. Why should we exploit one of these models?! 

This application can be created using Hugging Face, which provides a lot of models specialised in translations from one language to another. For example, you can pick this model that is targeting Italian translation from English. After you choose the model for the translation, you can concretize this idea by building an application with Streamlit.

Project link: Build your own AI translator   

4. Analyze research papers

During my research fellowship, I learned how to read papers rapidly and efficiently. But just reading a paper with a minimum of 30 pages is time-consuming and it’s hard to stay on top of research with this explosion of papers released every day. To boost research productivity, wouldn’t it be better to extract the relevant information from academic papers? These are the following three use cases that can be helpful for your career in the data science field.

Question & Answer over papers

Generating Questions and Answers from documents is one of the coolest applications that bring value. Most tutorials use Chat-GPT to create an automatic Q&A session, but it’s not the only solution. You can also create your personalised bot using LangChain and Sentence Transformers from HuggingFace. There are the following steps:

1. Load the PDF document using PyPDFLoader
2. Extract chunks from the text 
3. Extract embeddings using the Sentence Transformer library
4. Build the bot to answer questions 

Project links: 

• Question & Answer over papers with Chat-GPT API

• Question & Answer with LangChain and Sentence Transformers

Summarize papers

Another common use case is to summarize the paper. Like before, this task can be automated with generative AI tools. A cute web application can be built using GPT-3, LangChain and Streamlit.

Project link: Summarize papers

Query multiple papers

If we summarize multiple papers at the same time, it would be nice to filter to query these summaries based on questions. Wouldn’t it be cool? It can be again very simple by utilizing LangChain and OpenAPI-API.

Project link: Query multiple papers

5. Creating Code documentation

While working as a data scientist in my last experience, I have noticed how it’s important to document the code day by day. If you work by yourself, you don’t care about it. But when you work with a team, it becomes complex to manage tasks without code documentation. In particular, it can happen that a team member leaves the company and was the only person that understood his/her code. Even if documentation is really useful, it’s a very boring task that consumes time. Thanks to the boom of Large Language Models, we can again avoid this hard work by creating Python Docstring with Chat-GPT.

Project link: Creating Code documentation 

6. Automate Powerpoint presentations

If you are a data scientist, it surely happened that you had to prepare PowerPoint slides for discussing the results with the client. This is another time-consuming work that can be automatized thanks to Generative AI. You can ask Bing Chat to generate VBA code to create PowerPoint slides by specifying clearly the context and the information for each slide. 

Project link: Automate Powerpoint presentations

7. Sentiment Analysis of Reviews

In the world of industry, sentiment analysis of product reviews can help companies to understand if customers are liking or not products, allowing them to improve the service and stay competitive in the market. This is the classic data science project that requires a lot of steps to be solved: text-preprocessing, word embeddings and application of a machine learning model. 

The first step is the most laborious task, which requires a good understanding of the language you are analysing. This problem can be managed fastly by using Chat-GPT. In addition to this analysis, it can be possible to generate a list of pros and cons from each review, create a list of possible suggestions to improve the product and so on.  

Project link: Sentiment Analysis of Reviews

Final thoughts

That’s it! These are seven Generative AI projects that can help you to boost your resume and improve your efficiency at work. I suggest you try to have fun while working on the projects. Driven by inspiration, everything is possible. If you have an idea, try to make it in practice and voilà, you’ll be satisfied with the final product. Thanks for reading. Have a nice day!
 
 
Eugenia Anello is currently a research fellow at the Department of Information Engineering of the University of Padova, Italy. Her research project is focused on Continual Learning combined with Anomaly Detection.