Arufa|@arufa

"Make Exist First, then Improve it recursivly"

@arufadesuwa

@Radit

@404

@arufadesuwa

I'm Raditya, a vocational student at SMKN 4 Payakumbuh who fell for the maths behind machine learning and never really looked back.

I learn best by taking things apart. When something feels like a black box, I'd rather rebuild it from scratch than trust it blindly — writing my own backprop, my own optimizers, my own little libraries just to see the gears turn. It's slower, but nothing sticks quite like understanding you had to earn.

What keeps me going is the space between low-level systems and the models that run on them — Rust when I need the machine to be honest with me, Python when I need to move fast. I'm curious, a little stubborn, and happiest when I'm mid-way through building something I didn't fully know how to build yet.

Experience

January 2026 · June 2026

AI Engineer Intern

PT Mbot Technology Innovate

Focused on optimizing and fine-tuning LLMs for production. Reduced cloud inference costs by 30% through post-training quantization and structural pruning on internal LLM architectures, while maintaining a 95% accuracy threshold. Improved model F1-score from 0.78 to 0.92 by fine-tuning open-source LLMs with LoRA (Low-Rank Adaptation) on specialized internal domain documentation.

August 2025 · Present

FullStuck Dev & ML Head Division

TEFA SMKN 4 Payakumbuh

Head of ML Division and full-stack dev at the school's tech business unit. On the ML side, I own the full pipeline end-to-end — data pipeline, model experimentation, all the way to deployment. On the dev side, the job is designing a system architecture that can actually handle the model without making everything fall apart under load. Because a great model means nothing if the infrastructure isn't ready to receive its output.