---
title: "Muhammad Ariel Shakaramiro (@arielshakaramiro) | Hashnode"
description: "Newbie for AI Learning | Builder | Developer | AI Enthusiast. Read the latest articles by Muhammad Ariel Shakaramiro on Hashnode."
url: "https://hashnode.com/@arielshakaramiro"
---

@arielshakaramiro

# Muhammad Ariel Shakaramiro

@arielshakaramiro · Joined August 2026

Newbie for AI Learning | Builder | Developer | AI Enthusiast

### [One Answer, Four Different Scores: Learning LLM Evaluation the Hard Way](https://shaka-ai.hashnode.dev/llm-evaluation-offline-online-metrics-en?utm_source=hashnode&utm_medium=feed)

Hands-on LLM evaluation: EM, F1, BLEU, ROUGE, cosine similarity, plus a real online metrics dashboard built with FastAPI, MySQL, and Streamlit.

8 min read · [shaka-ai.hashnode.dev](https://shaka-ai.hashnode.dev/llm-evaluation-offline-online-metrics-en?utm_source=hashnode&utm_medium=feed)

[Sep 25](https://hashnode.com/posts/llm-evaluation-offline-online-metrics-en/6ab69604b85288396efeb78c)

### [Satu Jawaban, Empat Angka Berbeda: Belajar Evaluasi LLM dari Nol](https://shaka-ai.hashnode.dev/evaluasi-metrik-llm-offline-online?utm_source=hashnode&utm_medium=feed)

Belajar evaluasi LLM dari nol: EM, F1, BLEU, ROUGE, cosine similarity, dan dashboard online metrics pakai FastAPI + MySQL + Streamlit.

7 min read · [shaka-ai.hashnode.dev](https://shaka-ai.hashnode.dev/evaluasi-metrik-llm-offline-online?utm_source=hashnode&utm_medium=feed)

[Sep 25](https://hashnode.com/posts/evaluasi-metrik-llm-offline-online/6ab6916ea258136cc7500724)

### [Serving Your Own Model with llama.cpp: GPU Offload, Caching, and the Trap of a Stuck Wrong Answer](https://shaka-ai.hashnode.dev/serving-llm-llamacpp-cache-wrong-answer-trap?utm_source=hashnode&utm_medium=feed)

Serving a GGUF model with llama.cpp, caching answers in SQL, and a demo of why a cached wrong answer doesn't just go away on its own.

7 min read · [shaka-ai.hashnode.dev](https://shaka-ai.hashnode.dev/serving-llm-llamacpp-cache-wrong-answer-trap?utm_source=hashnode&utm_medium=feed)

[Sep 25](https://hashnode.com/posts/serving-llm-llamacpp-cache-wrong-answer-trap/6ab68b524750c7424a5bd7aa)

### [Nge-serve Model Sendiri dengan llama.cpp: GPU Offload, Cache, dan Jebakan Jawaban yang Nyangkut](https://shaka-ai.hashnode.dev/serving-llm-llamacpp-cache-jawaban-salah?utm_source=hashnode&utm_medium=feed)

Menyajikan model GGUF pakai llama.cpp, cache jawaban di SQL, dan demo kenapa jawaban salah yang tersimpan tidak bisa hilang begitu saja.

6 min read · [shaka-ai.hashnode.dev](https://shaka-ai.hashnode.dev/serving-llm-llamacpp-cache-jawaban-salah?utm_source=hashnode&utm_medium=feed)

[Sep 25](https://hashnode.com/posts/serving-llm-llamacpp-cache-jawaban-salah/6ab68ab0d4d10e9e6a0b69ad)

### [What Does Fine-Tuning Actually Learn? Testing QLoRA with Deliberately Controlled Data](https://shaka-ai.hashnode.dev/what-does-llm-fine-tuning-actually-learn?utm_source=hashnode&utm_medium=feed)

A controlled QLoRA experiment: format is learned first, facts come later, and brand drift turns out to just be about epoch count.

7 min read · [shaka-ai.hashnode.dev](https://shaka-ai.hashnode.dev/what-does-llm-fine-tuning-actually-learn?utm_source=hashnode&utm_medium=feed)

[Sep 25](https://hashnode.com/posts/what-does-llm-fine-tuning-actually-learn/6ab685bad8756b66d8d3aa49)