Skip to content
Personal projects

01 / Local intelligence

Turkish language model & Linux monitoring

Jazari is my Turkish model adaptation; ailm puts local AI to use in Linux monitoring.

Published model & open-source tool · Updated · September 2026

How the pieces connect
Linux events
Jazari 4B
Classification
Useful summary

What I worked on

Training experiments, evaluation and integration of a local model into a system-monitoring workflow.

The starting point

Training a model is only part of the work. The useful question is where a small, local model can help with a specific recurring task.

How I built it

  1. Adapt Qwen3.5-4B to Turkish with continued pretraining and supervised fine-tuning. The published training run used an RTX 5090 on vast.ai.

  2. Connect the model to ailm, which collects Linux events and uses a local LLM to summarize them and suggest what deserves attention.

  3. Publish the model card with evaluation results and limitations, including areas where performance remains limited.

The result

The model weights and monitoring tool are public. Together they show a path from a training experiment to a concrete local application.

Model-card benchmarks describe specific test sets, not universal accuracy. Suggested causes and actions still need verification. The earlier voice-coach experiment is a separate project.

Code & resources

Qwen / Fine-tuning / Python / Ollama / SQLite / Linux