Preparing page.

|

|

(00—04) / Personal field notesBandung, Indonesia

AFRIZAL Yafi.

[00] / The person behind the work

From understanding
patterns to building
practical things.

I work across machine learning, computer vision, and full-stack development, building things that are practical, thoughtful, and useful.
its mee
FIG. 01 / Outside the editorAY ↗
Research.
Software.
A work in progress.Explore the notes
[01] / An investigative streakComputer vision · Deep learning

Teaching machinesto see.

01 / The material

Learning to see.

I co-authored a computer-vision study using EfficientNetB0 transfer learning to classify 7,356 dermatoscopic images across five categories.

Normal / Melanoma / Arsenic / Psoriasis / Eczema
02 / The experiment

Question the model.

The work explored data augmentation, L1/L2 regularization, and dropout to mitigate overfitting, comparing training runs of 20, 30, and 50 epochs.

Transfer learning → evaluation → refinement
03 / The result

Look beyond accuracy.

The best model reached 89.12% validation accuracy at 50 epochs. Confusion-matrix analysis revealed class-level errors, including confusion between visually similar Psoriasis and Eczema samples.

Best validation result / 50 epochs
[02] / Putting it into practiceExperience & selected projects

From the model.To the real world.

Different scales.
The same attention.

My experience spans frontend work for companies and full-stack projects — from connected activity data to interactive digital spaces.

2023
Internship / July — August 2023

FAB Digital

Frontend Web Developer

Contributed to a website for PT Telkom Indonesia at FAB Digital, part of PT Telkom Indonesia.

React / Vite / Bootstrap
2024
Freelance / February — August 2024

PELOSA

Frontend Web Developer

Developed and maintained a website for a local company.

Astro / Tailwind CSS / AlpineJS

Selected builds.

03 project notes / Open to explore
/01ACTEData into a daily picture

A full-stack activity dashboard that retrieves, processes, and visualizes workout data. Strava OAuth2, activity webhooks, and rate limiting connect external services to a Cloudflare Workers and D1 backend.

Time-based matching brings workout records and heart-rate data together across multiple sources.

TypeScript / Astro / React / Workers / D1 / Vitest
/02Gadjah DudukA gallery with depth

A full-stack digital gallery with server-side rendering on Cloudflare Workers and an interactive WebGL experience built with Three.js, React Three Fiber, and GSAP.

Separate gallery, admin, and Sanity Studio applications share reusable packages. The admin includes password hashing, secure cookie sessions, protected routes, and image uploads.

React / TypeScript / Three.js / GSAP / Sanity / Workers
/03TasbhMateA small, purposeful tool

A responsive digital prayer counter built and deployed with Next.js, React, and TypeScript.

Multiple input methods, light and dark themes, and local-storage persistence keep the interaction useful across visits.

Next.js / React / TypeScript / Tailwind CSS / shadcn/ui
View the project index
[03] / A growing vocabularyTools, foundations & learning

A foot inboth worlds.

/01

Models

Machine learning & computer vision

  • Python
  • TensorFlow / Keras
  • EfficientNetB0 / CNNs
  • Transfer learning
  • Data augmentation
  • Model evaluation
/02

Systems

Software & infrastructure

  • JavaScript / TypeScript
  • React / Next.js / Astro
  • Cloudflare Workers / D1
  • REST APIs / Webhooks
  • OAuth2
  • Git / GitHub / Vite
/03

Workflow

AI-assisted engineering

  • Model Context Protocol
  • OpenCode / Claude Code
  • Structured prompting
  • Training / validation analysis
  • Confusion matrices
  • Overfitting analysis
Behind the work

Curious by nature. Deliberate by design.

I’m interested in where software engineering, machine learning, and thoughtful interfaces meet.

  1. 01

    Understand

    Start with the problem, the people, and the constraints.

  2. 02

    Experiment

    Explore possibilities, build something tangible, and test the assumptions.

  3. 03

    Refine

    Revisit the details—from how it works to how it feels.

Beyond the code

My experience also extends to public relations and licensing, working with marketing and admissions teams to connect with prospective students.

[04] / The directionStill exploring. Still building.

I’m interested in building practical AI/ML systems — and integrating models into reliable software products.

What’s next?

Let’s start a conversation.
Afrizal Ihyauddin Yafi
AI/ML & Software EngineeringLinkedIn ↗Back to the beginning ↑(END OF NOTES)