Skip to content

Projects

Things I've built

Applied work — systems, models, and tools. Peer-reviewed research lives on the Research page.

ScreenAI Lab
NLP Mar 2026 – Jul 2026 ★

ScreenAI Lab

Independently built recruitment platform supporting 312 registered users and 248 submitted applications, with rubric-based LLM scoring, identity masking, document-grounded evidence, and recruiter review.

PythonFastAPIReactIndoBERT +6
KCIC Data Automation
Data Engineering Sep–Oct 2026 · Yokke internship

KCIC Data Automation

An Excel/VBA workflow built during my Yokke internship to prepare reports from three banks, generate four standardized CSV outputs, and support Oracle ingestion with validation, logging, and archiving.

VBAMicrosoft ExcelOracle DatabaseSQL +2
LabLink
Full-stack 2026 · 3-month build ★

LabLink

A centralized internal system for MBC Laboratory — members, inventory, on-call rosters, attendance, and a transparent cash ledger — replacing a sprawl of scattered docs, drives, and spreadsheets.

PythonFastAPIReactVite +6
Flood Impact Segmentation
Computer Vision 2026 · national competition ★

Flood Impact Segmentation

A 10-class Swin Transformer + FPN model that maps flood impact from aerial imagery, built for the national Dolanan Data NEXUS 2026 competition — a Top-10 finalist entry.

PythonPyTorchtimmSwin Transformer +3
Big Data Toolkit
Deployment & Infra 2026 · study-group tool

Big Data Toolkit

Scaling and deploying MBC Lab's no-code ML-lifecycle teaching toolkit — a Streamlit app that collapsed past one user on a free tier — onto a GCP VM that served the whole study group.

PythonStreamlitDockerGCP Compute Engine +1
Operational Risk Analysis
Machine Learning 2026 · course project

Operational Risk Analysis

XGBoost model that predicts job-level risk (completed / delayed / failed) on a hybrid manufacturing line, with SHAP turning predictions into actionable maintenance and scheduling insight.

PythonXGBoostSHAPscikit-learn +2
Indonesian Tweet Emotion Classification
NLP 2026 · course project

Indonesian Tweet Emotion Classification

A comparative text-mining study classifying Indonesian tweets into five emotions, benchmarking a BiLSTM baseline against IndoBERT and measuring how much custom preprocessing actually moves the needle.

PythonPyTorchHugging Face TransformersIndoBERT +2