Learning Software Development is hard. Especially when you can't understand the lecturer's presentations. Wouldn't it be nice to have someone explain them for you?
You are going to implement a Python script that explains Powerpoint presentations using the GPT-3.5 AI model. The script will take as input a presentation file, send the text from each slide to GPT, and save the results together in an output file.
Cool, right?
Your script should do the following:
- Take the path of a
.pptxfile. - Parse the presentation to get its data.
- Go through every slide separately, and:
- Extract all the text from all text boxes.
- Insert the text into an appropriate prompt for GPT.
- Send the prompt in a request to the OpenAI API.
- Extract the AI's reply from the response.
- Gather the explanations for all the slides in a list.
- Save the list in a JSON file.
It takes time for the GPT models to generate responses. If we send the slides one after another, we will have to wait for a very long time. Therefore, we will make asynchronous API calls, so that all the slides are processed at the same time.
Use Python's async/await syntax, together with the builtin asyncio package.
Tip: Here is an awesome guide.
- Your code should ignore slides without text.
- Your code should handle weird whitespaces within presentation text.
- The name of the output file should be the same as the original presentation (but with a
.jsonsuffix).
- Add a timeout to the requests you make to the OpenAI API.
- Add a convenient CLI interface for your script (use
argparse). - Handle errors raised while processing a slide. Instead of allowing the entire program to crash, save the explanations for the other slides as usual, and put an informative error message for the slide that failed.
Tip: Here is a list of potential errors you might encounter.
There are several Python packages that can help you parse .pptx files and extract their data.
We recommend installing the python-pptx package.
Tip: The documentation of
python-pptxis a bit annoying. Concentrate on the parts that are relevant for what you need.
To send requests you will need to create an OpenAI account on the OpenAI website. There is a "Sign up" button on the top-right.
You will then need to generate an API key, that will be your identifier when using the API.
Also, instead of sending HTTP requests directly, it is much easier to use the official openai Python package to do that for you.
Tip: Here is a good guide that covers everything.
OpenAI is a company that develops AI models. Each model specializes in something different. For this project, it is best to use the gpt-3.5-turbo model (it's also the model used by the ChatGPT website).
You should think carefully how to ask GPT in a way that will give you the best answers. You cannot just throw some text from a presentation and expect it to know what you want. Here are a few good questions to ask yourself when designing a prompt:
- How can I clearly explain what I want the AI to do for me?
- Which information is relevant besides the text of the slide?
- Is there any additional background that might be useful for the AI to know?
Tip: You can easily test your prompts on ChatGPT.
When you create a new OpenAI account you will be able to use the API for free for 3 weeks, or until you use a certain amount of tokens. After that you must pay. You can check how much you've spent on your account page.
Don't worry! You should have enough free tokens so you don't have to pay during this project. After 3 weeks you can create a new account with a different email (and phone number), and generate a new API key.
Also, free users have a rate limit of 3 requests per minute. If you try to send 20 requests at the same time, don't be surprised if you get errors...
Your work will be graded based on the following criteria:
- The code passes all mandatory requirements.
- Code quality
- Readable code
- Indicative names
- Spacing and indentation
- Avioding duplicate code
- Comments and documentation
- Proper division of logic into small functions
- Proper division of modules into separate files
- Proper usage of Git
- Small and standalone commits
- Descriptive commit messages
- Working on different features in separate branches
- Creating a proper pull request
- Not uploading junk files that are not part of the code (use
.gitignore)
- Bonus requirements you may have implemented
Tip: Here are some guidelines on good commit practices.
- Before you write code:
- Make sure you have a vision for the flow and architecture of your project.
- Play a bit with the different packages. Not too much. Just to feel comfortable.
- Ask yourself the following:
- Which logical parts does my project have?
- Which files should I have?
- Which functions should I write?
- Which branches should I open?
- Which parts of code will depend on others? Which won't?
- Which features are central? Which are less important?
- Good questions to ask while coding:
- Do these changes really belong in the same commit?
- Can I give a name to this piece of logic and make a separate function?
- Do I really have to fix this issue right now?
- Is this really worth the time I'm investing?
- If I want to add extra features later, will it be easy?
- If someone else looks at this code, will they hate me?
- Will I understand this commit message in a week from now?
- Before you submit:
- Code-review yourself! It's a good practice.
