LLM-Assisted Development

A large language model (LLM) is a type of artificial intelligence (AI) program capable of recognizing, understanding, and generating text, among other tasks. LLMs are trained on vast datasets using machine learning, specifically deep learning techniques built on a neural network architecture called a transformer model (Cloudflare, 2022).

LLMs are highly flexible and versatile. A single model can perform a wide range of tasks, including answering questions, summarizing documents, translating languages, and completing sentences. While not perfect, LLMs demonstrate a remarkable ability to make predictions and generate meaningful responses from a relatively small number of prompts or inputs. They can also be used for generative AI, producing new content in human language based on input prompts (Amazon Web Services, 2025). Common applications include:

Knowledge base answering, often referred to as knowledge-intensive natural language processing (KI-NLP)
Text classification
Code generation
Text generation

Generative AI refers to algorithms, such as ChatGPT, that can create new content, including audio, code, images, text, simulations, and videos (McKinsey & Company, 2024).

During the development of the Clubs Connect project, the team leveraged LLMs to assist in completing tasks efficiently. The following are personal reflections from team members on how they used LLMs, which models they used, and their individual perspectives on the benefits and limitations of these tools.

“While working on Clubs Connect, I used Large Language Models (LLMs), mainly ChatGPT and DeepSeek, to assist me with different parts of the development process. I used it to generate code snippets, fix bugs, and understand how certain features should be implemented. Although I did not always fully grasp how all the generated code worked at first, using the LLM helped me identify patterns, improve my problem-solving. I often went back to study or test the code to understand it better. It helped me move forward when I got stuck and made it easier to see how certain parts of the app could be implemented. The LLM was also helpful for explaining error messages and suggesting more efficient or readable code structures. Overall, the LLM was a valuable tool that guided me through challenges and allowed me to progress faster, but it also reminded me of the importance of building a stronger understanding of the underlying concepts and practicing on my own to build a strong foundation in coding.” Paballo Molaontoa

“I used large language models like ChatGPT and Claude mainly for debugging and troubleshooting my code. They were helpful in quickly spotting bugs that would’ve taken me a long time to find by going through hundreds or even thousands of lines manually. Using them saved a lot of time and made the debugging process much smoother. Besides just fixing errors, I also used them to get explanations of certain programming concepts and to see different ways of solving a problem. This helped me understand my code better and improve how I write it. Overall, the LLMs became useful tools in my workflow, especially when I needed quick feedback or a second perspective on a problem.” Mmakwena Moichela

“During the development of the Clubs Connect application, large language models were an integral part of my contribution to the project. I mainly used ChatGPT, an OpenAI language model, and occasionally Claude by Anthropic when I ran out of credits, as I used the free version of ChatGPT. I used LLMs for different things, they helped me a lot with getting things done on time and being able to do the work quickly so that I could have time for my other three courses. I used them for code generation, brainstorming documentation ideas, determining what to include and how to structure it, and as the team member responsible for documentation, I also used them for grammar checks and sentence refinement to ensure our documentation maintained a professional tone. In most cases, about 99% of the time, the LLMs provided useful, accurate, and timely assistance. However, there were instances where I had to clarify my prompts or manually refine the responses, especially when the generated tone did not match my intent or when the code required adjustments to function as I needed it to. Personally, I think LLMs are useful as a tool especially when there is so much to do with little time to learn and put in time to figure things out from scratch. Do I think that LLMs make one lazy and not learn as they should, yes. As Andrew, a correspondent from TIME referenced Natalia Kosmyna, who led an MIT study on the impact of LLMs, “even if efficiency increases, reliance on AI can reduce critical thinking, creativity, and problem-solving” (Chow, 2025). I fully agree with this perspective. Having used LLMs for nearly two years, I have observed that passive or excessive use leads to laziness and over-reliance and instead of helping with learning, it enables procrastination and ultimately impedes learning. As Rus and Kendeou (2025) note, LLMs constrain the very learning processes necessary for learning to occur: reflection, analysis, evaluation, application, transfer, and synthesis. I am aware that LLMs have more negatives than what meets the eye but at the same time I need to get my degree, its been long overdue, and LLMs have been getting me through the Wits edge, now I have the dilemma to develop the discipline and integrity when I enter the workplace so that I can actually learn and gain the skills instead of prompting my way to a paycheck and not growing as I should.” Busisiwe Mnisi

“For our project, I made use of Large Language Models (LLMs) to assist me in understanding, diagnosing, and correcting technical errors encountered during development. Whenever I ran into runtime errors, CORS issues, or API request failures, the LLM provided step-by-step explanations of the error messages and helped me identify the underlying causes. This guidance significantly accelerated the debugging process, reduced trial-and-error efforts, and allowed me to focus on implementing effective solutions rather than spending excessive time figuring out what went wrong. Beyond debugging, I also used the LLM to help translate my ideas and design decisions into working code. For example, when I wanted a button to appear only for users with a specific role, the LLM guided me in writing conditional rendering logic in React, showing me how to structure the code safely and efficiently. It acted like a real-time programming assistant, helping me think through implementation details and anticipate potential issues before writing code. Additionally, I leveraged the LLM to generate clear and structured explanations of errors and solutions. I specifically used ChatGPT (GPT-5-mini), for debugging complex front-end and back-end interactions, understanding configuration and network issues, and providing actionable guidance that improved both development efficiency and the overall quality of our project deliverables.” Theto Maunatlala

“While working on Clubs Connect was limited to the Google AI Overview which uses an LMM engine but works differently. The reason for my limited use of LLM’s is because I wanted to use this project as an opportunity to learn and I feel that as an individual I learn more when I do the work myself rather than using an LLM. An LLM would provide a shortcut and in moments of pressure it would encourage me to focus on getting things done rather than putting in the time. I feel that at this stage of my career efficiency at the cost of learning would not be beneficial for me so I needed to utilise this moment of practise. I chose this position because of it allows me to develop my problem-solving skills by forcing me to think through my problem and find a way to implement it. It has also helped me learn how to accurately describe problems when I encounter them and how to find solutions that will work for me. Knowing where and how to find solutions is also a skill. I feel that LLM’s will be more beneficial for me after I have solidified my foundation and I know what I am doing.” Mukondeleli Ramabulana