codeLynn - fantasies and MLOps

Ehem, I am here, in the cold light of the day~

Now I am in the stage that studying on Coursera feels bland and isn’t impressive anymore.

I heard the term “MLOps” on Threads and it stirred my mind in the weekend. Then now, I decide to make an AI chatbot product, serving for my needs.

Unlike Graph-of-Models which is focus on researching and propose method, codeLynn focuses on bringing the art into deployment.

Definition of codeLynn

I want to create a virtual self of mine. You know c.ai, yes, I want to do something similar to this :> But it will have my tone and my personalities.

Seems scary and funny at the same time, hopefully it won’t be unhinged like me.

Why I use this? Maybe just need a friend shares the same brain frequency. Sometimes, there are thoughts and ideas I don’t dare to share to human, because it can be triggering.

I want to be open more about myself, and nothing better than an AI agent that accumulate more knowledge than mine, know about me and don’t judge me.

It should be an AI system have general knowledge that have fine-tuned modules in coding, mathematics and languages like German or Latin.

Its tone and persona should be like me. This part maybe will be RLHF (Ouyang et al., 2022). The chatbot will be text-focused, I don’t think about image generation or voice chat.

What is MLOps?

MLOps is a paradigm, an engineering practice that is a mix of ML, software engineering and data engineering. It’s the procedure to make our ML models from development stage get into production in a consistent and reliable manner (Stone et al., 2025).

In a full cycle of MLOps, there are a lot of stages like image below. In reality we won’t need to implement some steps as our scale is far smaller.

MLOps lifecycle (Ouyang et al., 2022).

My work will use Kubeflow as the core spine.

Kubeflow architecture (Source: Kubeflow).

I will update in the next post more things about datasets and models