Portfolio — Vol. 01Informatics graduateYogyakarta  WIB

Azrial Ahmad Haidar

Looking up through a market hall at a blue glass skylight hung with banners
Pl. I From the street series

Exploring the intersection of data, AI, and web — building things that make sense of complexity. Also happens to make photos.

View work GitHub
  • Data Engineering
  • Machine Learning
  • Computer Vision
  • Quality Assurance
  • Full-Stack
  • Street Photography
  • Cosplay Portraits
  • Editorial Design

01About

Curious by nature,
precise by practice.

I’m a fresh Informatics graduate from Indonesia who finds himself equally at home writing a data pipeline, training a detection model, or building a web interface. I don’t pick one lane — I follow the problem wherever it leads.

When I’m not in front of a terminal, I’m out with a camera looking for candid moments in the street, or setting up portraits for cosplay photography. I believe the same attention to detail that makes a good photograph makes good software.

02Projects

Selected work

All on GitHub
  1. mic + calltranscript0:040:110:180:250:320:39notesID · ENlocal · whisper.cpp
    Fig. 1 Listen, transcribe, write up

    fivesight

    A meeting notepad for Windows that listens on your own PC. Speech is transcribed locally with whisper.cpp on the GPU; your shorthand is expanded into notes, and Ask answers questions about meetings, your files, and Jira — read-only, every command checked against an exact allowlist. Works in Indonesian, English, or both.

    • Rust
    • Svelte
    • TypeScript
    • whisper.cpp
    • Jira
    Source
  2. 400422PIIguardFlash70Bjudgesimple → fast modelcomplex → large model
    Fig. 2 Governance pipeline

    ai-governance-proxy

    Gatekeeper, an inline proxy between apps and LLMs: scrubs PII with regex and a local model, blocks prompt injection before any paid call, routes by complexity between Gemini 2.5 Flash and Llama 3.3 70B to save budget, and rejects answers an LLM judge scores as likely hallucinated.

    • Python
    • Ollama
    • Gemini
    • Groq
    • LLM-as-a-Judge
    Source
  3. 123circle 0.97YOLOv11 · ONNX
    Fig. 3 Detection loop

    osu-aim-assistant

    An AI-powered aim assist tool for osu! using YOLOv11 for real-time target detection and high-performance screen capture. Explores the boundary between computer vision and game interaction.

    • Python
    • YOLOv11
    • OpenCV
    • ONNX
    • Computer Vision
    Source
  4. n = 90+
    Fig. 4 Composition study

    virtual-photography-analysis

    A data analysis project that explores a dataset of 90+ top-rated virtual photos from gaming to uncover the key visual and compositional characteristics behind a successful gaming photograph.

    • Python
    • Pandas
    • Jupyter
    • Visualization
    Source
  5. 01 baseline0203 biomechanical
    Fig. 5 Three trajectories

    mouse-movements

    An interactive, side-by-side visualizer for three mouse movement algorithms commonly used in aimbots — from a simple linear baseline to a full biomechanical model.

    • Vite
    • JavaScript
    • Pico CSS
    Live demo Source

IndexAlso on the bench03 repos

03Publications

Research & writing

IEEE Xplore10th International Conference on Information Technology and Digital Application (ICITDA) · 2025

A Comparative Analysis of YOLOv12 Model Sizes for Road Damage Detection on an Indonesian Dataset

Azrial Ahmad Haidar Daffi Zuhri, Dhomas Hatta Fudholi, Rahadian Kurniawan, Tisara Sita

Read paper

04Photography

Through the lens

I shoot street and candid — moments as they are, unposed and honest. From time to time, the subject is a cosplay portrait: a different kind of realness.

Series AStreet & Candid07 frames

Series BCosplay Portraits06 frames

05Design

Visual thinking

Design is how I make sense of things before writing a single line of code. Layouts, editorial compositions, and interface concepts — a quieter side of the same curiosity.