DevDynamo | Devpost

Inspiration:

I'm not an expert in AI, but I'm always curious about how technology can make my work easier. I often read news and watch videos about AI. One day, I thought, "Why not use AI to help me with my coding tasks?" That's where the idea for this project came from.

Learning Journey:

I don't have formal training in AI, but I try to learn as much as I can by reading and watching simple explanations. I experiment with AI tools when I can, even though it's challenging sometimes. But I believe in learning by doing, so I keep trying.

How I built it:

The project started with a simple idea: using AI to automate tasks in Linux. I built this program with Python, utilizing the Gemini AI API. The logic behind it is that I trained the model to interpret commands in natural language and convert them into Python code.

To achieve this, I divided the dataset into two parts. One part is responsible for executing operations, such as opening applications or performing specific tasks like downloading files, while the other part focuses on error handling.

For instance, when I instruct the model to "open Google," it generates the Python code necessary to perform that operation and sends it back with a designated tag, "#code". My program then detects and extracts the code following this tag, executing it accordingly.

Moreover, if an error occurs during execution, such as a package installation issue or a syntax error in the generated code, the program seamlessly transitions to the error-handling dataset. Here, it attempts to diagnose the issue and provide a solution automatically, ensuring a smoother user experience.

This approach allows for a more robust and comprehensive AI assistant, capable of not only executing commands efficiently but also addressing potential errors that may arise during the process.

Challenges I ran into:

There were many challenges, especially because I'm not an expert in AI. Understanding how the APIs work and integrating them into my project was tough. I had to ask for help sometimes and spend a lot of time researching. But I didn't give up, and eventually, I made progress.

Accomplishments that I'm proud of:

Even though I'm not an expert, I'm proud of what I've achieved with DevDynamo. Building it from scratch was a big challenge, especially with my limited knowledge of AI.

I'm proud that I managed to make DevDynamo work with Gemini AI, even though it was tough to figure out at first. Using Python, a language I'm comfortable with, was a big help in getting things done.

I'm also proud that I focused on making DevDynamo helpful for other developers like me. I wanted to create something that makes coding tasks easier, and I think DevDynamo is a good start.

While there's still a lot to improve, I'm happy with how far DevDynamo has come. It shows that with determination and some learning along the way, you can create something useful.

What's next for DevDynamo:

While the current version is just a demo and has some bugs, I'm dedicated to making it better over time. In upcoming updates, I'll add more features to help tackle larger projects like making websites or games.

One of the big things I'll add is the ability to handle complex projects better. I'll train DevDynamo with new datasets so it can understand and break down big projects into smaller parts. This will make it easier to work on these projects step by step.

To make sure DevDynamo's code is solid and reliable, I'll also test it thoroughly using special datasets. These tests will help catch and fix any bugs, making DevDynamo work smoother.

I also want DevDynamo to understand your computer better. So, I'll add features to gather details like usernames or specific settings from your computer. This will help DevDynamo create code that fits your computer perfectly.

And I'm really excited about connecting DevDynamo with Gemini Vision. With this, DevDynamo can take screenshots of your screen and use Gemini Vision to understand what's on it. For example, if you say "Gemini, solve this problem," DevDynamo will take a picture of the problem on your screen and send it to Gemini Vision. Gemini Vision will figure out what the problem is, like an error message or code, and then DevDynamo can help fix it.

Additionally, I'll handle errors that can be fixed using a Graphical User Interface (GUI). For this, the model will use packages like pyautogui, which can simulate mouse movements and keyboard inputs. This will allow DevDynamo to interact with GUI elements on your computer to resolve certain types of errors automatically.

By adding these features, DevDynamo will become even better at helping you with your coding projects and making your life easier.