AI/ML engineer · Banda Aceh, IDAI/ML engineer · Computer vision · Web
ZidanMubarak
I fine-tune vision-language models until they can read Jawi handwriting.
01About
I work where language models meet handwriting.
A degree in Information Technology at UIN Ar-Raniry, Banda Aceh, ended with a thesis on teaching a vision-language model to read Jawi, the Arab-Malay script the base model read at 82.5% character error. Getting that down to 17.4% took three data scenarios, a failed epoch 4, and one decision to stop at epoch 3.
The unglamorous half of machine learning is the part worth keeping: data provenance, honest evaluation, and an interface someone can use without reading a README. Python for the models, TypeScript when the result needs to live in a browser, layouts designed by hand.
- Timezone
- UTC+7 (WIB)
- Focus
- Vision-language · OCR
- Now
- Open to AI/ML roles
02Services
Four ways I can help.
Not a menu of everything I have touched. The four areas where there is work below to prove it.
Fine-tuning vision-language models, training classifiers and recommenders, and evaluating them honestly. The work starts at dataset curation and ends with a checkpoint someone can call through an API or a demo.
- Vision-language fine-tuning with QLoRA
- Computer vision and transfer learning
- Recommendation systems
- Evaluation: CER, WER, accuracy
Evidence
- Jawi OCR
- Pacu Pintar
- NusantaraGo
Next.js and TypeScript for the front end, Flask and FastAPI when a model needs a door, and CodeIgniter where an existing PHP deployment still has to keep running. Typed data, accessible markup, deployments that fit on a free tier.
- Next.js, React and TypeScript
- Flask and FastAPI services
- MySQL and REST APIs
- Accessible, responsive front ends
Evidence
- Memory Vault AI
- NusantaraGo
- Guest Book System
Editorial layouts, dashboards and interface systems · including the ID-card and check-in flows in the guest book system and the charts that carry the argument in the climate analysis. Type, spacing and hierarchy treated as part of the engineering, not decoration on top.
- Interface and layout design
- Design systems and tokens
- Data visualisation
- Information hierarchy
Evidence
- Guest Book System
- Climate × Agriculture
Laboratory assistant for Algorithms and Programming and for Big Data Analytics at UIN Ar-Raniry. Java fundamentals in one room, OpenCV and CNN transfer learning in the other, plus the case studies and rubrics that made both practicums gradeable.
- Java programming fundamentals
- Computer vision practicum
- Case study and rubric design
- One-to-one debugging
Evidence
- Laboratory Assistant, Algorithms and Programming
- Laboratory Assistant, Big Data Analytics
03Selected work
Six projects worth your time
Chosen for impact and for how much of the evidence is public. The first one is the thesis.
Jawi OCR
2026 · ResearchCER 82.51% → 17.35% on 790 unseen images
Read the case study
Pacu Pintar
2026 · Machine learningRandom Forest · ≈90% accuracy
View project
Memory Vault AI
2026 · Open sourceFour memory types · local-first · on PyPI
View project
NusantaraGo
2025 · RecommendationTF-IDF + collaborative filtering · 107 commits
View project
Guest Book System
2025 · Web applicationQR check-in · generated ID cards · Excel export
View project
Climate × Agriculture
2025 · Data science10,002 rows × 15 columns · four metrics
View project04Experience
Record: experience, education and certifications
- Experience
- 05
- Education
- 01
- Certifications, award, paper
- 05
Entries and dates trace back to the CV and the public credential pages. Newest first; one open at a time.
Experience
05 entriesPracticum support for the computer-vision half of the Big Data Analytics lab.
- Assisted practicum sessions on image processing with Python and OpenCV, image classification with CNNs, and transfer learning with fine-tuning.
- Unblocked students on the material and on the technical failures that come with it · environment, tensors, training runs.
- OpenCV
- CNN
- Transfer learning
- Python
Java fundamentals practicum, plus the case studies and rubrics used to grade it.
- Guided students through Java basics: variables, data types, input, arithmetic operators, branching.
- Wrote practicum case studies at matched difficulty, each with its own grading rubric.
- Java
- Programming fundamentals
- Rubrics
Two-person ML team on Pacu Pintar, an AI learning-analytics platform. The capstone took Best Capstone Project.
- Built the machine learning component for Pacu Pintar, integrated with the team's Next.js and Hapi.js application.
- Trained a Random Forest classifier for three study-pace categories at ≈90% accuracy.
- Built the FastAPI ML API that served the classifier and LLM-based personal study suggestions.
- Completed the cohort's soft-skills track for work readiness.
- scikit-learn
- FastAPI
- LLM
- Teamwork
Six months of certification-based machine learning, ending in the NusantaraGo-ML capstone.
- Completed every class and certification project in the machine learning track, including the soft-skills and financial-literacy modules.
- Built an image classifier with a CNN and transfer learning.
- Top contributor on NusantaraGo-ML: a tourism recommender with content-based and collaborative filtering, served through Flask.
- CNN
- Transfer learning
- Recommender systems
- Flask
Three months digitising and reporting financial data for a provincial finance agency.
- Digitised and processed financial data with spreadsheets and analysis tools.
- Wrote internal documentation and daily operational reports, and supported the financial reporting workflow.
- Spreadsheets
- Data entry
- Documentation
Education
01 entryS1, Bachelor's degree
ThesisImplementasi Multimodal Large Language Model Untuk Ekstraksi dan Transliterasi Printed Jawi Script
- Graduated 2026.
- Thesis produced the public Jawi adapter and Gradio demo on Hugging Face.
Certifications, award, paper
05 entriesAsah led by Dicoding
Awarded for Pacu Pintar, the AI learning-analytics platform built by a two-person ML team inside a larger web team.
Dicoding Indonesia · Advanced
MLOps foundations: experiment metadata, model serving, monitoring and alerting.
- MLOps
- Model serving
- Monitoring
Academic Journal of Computer Science Research
Co-author (third of four). A Hebbian network recognises the Hijaiyyah letters waw and ba from 5×5 bipolar matrices, implemented and verified in Python.
- Neural networks
- Pattern recognition
- Python
Dicoding Indonesia · Advanced
Applied machine learning: predictive analytics, sentiment analysis, computer vision and recommender systems.
- Predictive analytics
- Computer vision
- Recommenders
Dicoding Indonesia · Intermediate
Preparing quality data for data-driven projects with Python, SQL and PostgreSQL.
- Data processing
- SQL
- PostgreSQL
05Tools
What I work with.
Tools that show up across these projects and the work behind them. Icons from simple-icons.
01
Machine learning
- Python
- PyTorch
- TensorFlow
- scikit-learn
- Hugging Face
- Keras
- OpenCV
02
Data
- pandas
- Jupyter
- NumPy
- SQL / PostgreSQL
- Plotly
- Streamlit
03
Web
- TypeScript
- Next.js
- React
- Tailwind CSS
- PHP
- CodeIgniter
- Bootstrap
- JavaScript
04
Deployment
- FastAPI
- Flask
- Gradio
- Docker
- Vercel
- PyPI
05
Tools
- Git
- GitHub
- GitHub Actions
- Linux
- Google Colab
06Contact
Let's makesomethinglegible.
Tell me what you're building. A sentence about the role, the project, or the problem is enough.
Elsewhere
- LinkedInin/zidan-mubarak (opens in a new tab)
- GitHub@zidanmubarak (opens in a new tab)
- Hugging Face@zidanmubarak (opens in a new tab)
The form replies from this address. Nothing is added to a list.