How I Approach & Think While Working on Projects in AI/ML

Hey, everyone. I'm Manikanta, and I'm currently pursuing an M.Tech in CSE.

As an engineering student, I’ve worked on many projects. When I say “many,” you might assume a number like 10, but what if I told you it’s much higher than that? I’ll dedicate a separate blog to discussing my projects in detail, but for now, let’s focus on how I think when working on projects.

How I Approach an ML/DL Project

During my B. Tech, I freelanced by working on projects for fellow students, which helped me build several unique and interesting projects. It was during my 3rd year, 2nd semester, after learning how to write research papers and prompt engineering- where we will give every instruction in a detailed manner in a sentence/paragraph to get the best responses from AI models. Then, I started gaining deeper insights into how models in ML and DL work. From that point, I made it a priority to absorb as much knowledge as possible, and this continuous learning helped me improve.

At first, my knowledge was limited to running projects on Jupyter Notebook and Google Colab. All I knew was

1. Give a prompt and run the cell.

2. If there's an error, copy it into ChatGPT, apply the corrected version, and run it again.

Initially, most of the projects I worked on were in machine learning, there weren’t too many problems with datasets or code. The major change in each project was the algorithm being used.

Gradually, my projects became more logical, while the students seeking project help had no understanding of their ideas. Once, a student came up with an Idea Trying to combine Data Augmentation and autoencoders, which finally led to an amazing experience later. I will try to explain how I tried to work with those. In the same way, they would come up with excellent topics but had no clue what they meant. This realization led me to change my approach. I decided to go beyond just providing prompts and instead truly understand and build projects from scratch.

Whenever I come up with an idea or receive a request for a project, I begin by mapping out the entire process:

• Defining the input, output, and overall workflow.

• Identifying elements that can be modified or ignored.

• Refine the idea at this stage to ensure a well-defined project scope.

Once the idea is structured, the next step is model selection. This is a critical phase, and the more time I invest here, the more I understand the problem at hand.

When working on a model, the first decision is choosing the right framework. We have various frameworks like TensorFlow, PyTorch, and others. I prefer TensorFlow (as a side note, I also hold a Developer Certificate in it). After selecting the framework, I focus on building the model while carefully considering all the main things I’ve outlined(What is my input, what are the steps we are following, what is the output at each step, what is my final output).

Why I Prefer TensorFlow

During my 3rd year, 2nd semester, there was a subject where all students who specialized in AI & IPA had to complete the TensorFlow Developer Certification. Since it was a requirement, I started learning TensorFlow. While preparing for the certification, I gained hands-on experience with:

• Efficiently using TensorFlow packages.

• Building models from scratch.

• Understanding deep learning workflows.

From the very beginning, I found TensorFlow comfortable and well-structured, which is why I prefer using it for my projects.

What’s Next?

In this blog, I have provided a brief overview of how I finalize a model. In the next one, I will explain how to build a complete project step by step.