Longer version.

I’m Mustafa. I’m an AI and backend developer — I build async Python services around LLM systems: multi-agent orchestration, RAG pipelines, structured output, and the data layers underneath. The evaluation that keeps them honest is usually part of the job too, because in practice those are not separable jobs.

I came to this from the web development side rather than the machine-learning side, but I had some idea of ML. I am more interested in the systems and the backend engineering side of AI. Most of what I’ve learned in the last two years is web development and integration of Generative AI. Stepping into generative AI has got me to explore this area more deeply, hence I started my journey towards AI engineering.

What I’m working on

Right now: the workspace operations copilot — a LangGraph orchestrator that routes between a Pinecone RAG knowledge base and a Supabase room-booking agent, with Instructor-validated tool calls and a bounded hop counter — and the evaluation harness that gates changes to it: sixty scenarios that pin the graph state and score how often the orchestrator claims to know something it doesn’t. The harness came second and turned out to be the more interesting project, because it measures a number almost nobody reports and the one users actually feel.

Alongside that I’m studying computer science, and doing a fair amount of the unglamorous work. I explore the dimensions of linux operating system part-time. I also use arch btw.

How I got here

I started by building things that worked in a demo and fell apart in use, which is a good teacher if you pay attention. I came to LLM systems from the services side, so I’ve always treated a model as one component in a pipeline rather than the pipeline itself. The turning point was the first time I had to explain to somebody my code and decisions, and I did not have a good answer. That took me understanding decision-making and tradeoffs while building. Now, I try to build things more mindfully and measuring decisions.

Terminal

I installed linux around summer 2025, since then I haven't looked back. The linux command line has been a great friend in this journey. My interest of the terminal made me learn vim motions, and now I use neovim as my main IDE. I use terminal for stuff outside of development, like managing my tasks, writing, etc.

What I’m looking for

Work around LLM systems where correctness matters more than demo speed: agent architectures that have to survive contact with real users, RAG and structured-output pipelines, evaluation for teams who suspect their numbers are flattering them, and the async Python services that hold it all together. If that is the shape of your problem, I would like to hear about it.