AI-Driven Pipeline Development
Date Posted: 08-26-2025
Date Last Updated: 09-01-2025
I have always been fascinated with local usage of LLMs, integrating local models to tool pipelines that can help assist the learning process for beginners or help assist in other development processes. While it is unfortunate that local models are very resource hungry, it is still quite interesting to come up with ways to utilize them as automation or assistance tools.
Here is a very simplified pipeline that in my opinion, has a lot of potential.
In the case of my team's thesis project, it all starts with an assumption of: "What does an LLM need to know before it infers that it will create a test case?". Surely it is not only a prompt that says "Create a test case", as an LLM may need more 'context' for that. That is where scraping technologies get included in the idea mix, to scrape website elements that will serve as context on what test cases to generate. It can be a straightforward pipeline on paper, but it opens the possibilities on other similar ideas or concepts. Such as automating project idea flow, or scaffold generation.
If more of us can use Local LLMs with more optimized/resource friendly algorithms, which is probably too ideal, I find myself very excited to dabble with some assitive software tool ideas. But for now, nope, as it is too resource demanding for my machine in particular for now. Strong local AI requires strong computing power.