Projects

A collection of digital products, experiments, and open-source contributions I've worked on.

ScamShield ID: Building an Evidence-Grounded AI Agent for Digital Fraud Triage

Merancang workflow AI agent berbasis Langflow dan Gemini untuk menganalisis bukti dugaan penipuan digital, menghasilkan structured risk assessment, mengekstrak entitas penting, dan menyusun rekomendasi berbasis bukti.

🤖 Built a multi-step AI agent workflow for structured digital fraud triage

🧩 Enforced schema-constrained LLM outputs for consistent machine-readable risk assessments

🔍 Designed evidence-grounded analysis to reduce unsupported conclusions and hallucinated case details

LangflowGeminiGenerative AILLMPrompt EngineeringStructured OutputsAI Agent WorkflowJSON Schema

Investigating Shortcut Learning in ASL Classification with Grad-CAM

Membedah fenomena Clever Hans Effect pada model CNN kustom, mengevaluasi limitasi arsitektur MLP, dan menganalisis kerapuhan model terhadap Data Distribution Shift di lingkungan inferensi dunia nyata.

🏆 100% accuracy on the held-out internal test set with an ultralight 135k parameter CNN

🔍 Exposed and mitigated dataset-specific background bias using targeted spatial augmentation

⚠️ Diagnosed critical model fragility and Data Distribution Shift in real-world inference using XAI

TensorFlowKerasOpenCVGrad-CAMtf.data APIScikit-LearnMatplotlibSeaborn

Credit Score Classification: Leakage-Aware Evaluation & Risk-Averse Deployment

A leakage-aware machine learning pipeline for classifying customer credit scores using group-based validation, automated preprocessing, and conservative deployment logic.

🎯 0.69 Macro F1 with XGBoost

🛡️ Customer-aware evaluation with StratifiedGroupKFold

⚖️ 35% Poor-risk override in Streamlit

Scikit-LearnXGBoostImbalanced-LearnStreamlitPython

Big Data Pipeline for Churn Prediction

Membangun distributed pipeline berbasis PySpark (Medallion Architecture) untuk membedah silent attrition dan limitasi model prediktif pada marketplace non-kontraktual.

🏆 Improved ROC-AUC by 15% using Random Under-Sampling

💾 Processed 100k+ imbalanced ecommerce transactions

🔍 Global & Local Explainability via SHAP Values

PySparkMedallion ArchitectureRandom Under SamplingLogistic RegressionRandom ForestXGBoostK-Fold Cross ValidationSHAP

Batara

A gamified language learning app prototype built in 30 hours for Garuda Hacks 6.0, featuring Generative AI storytelling.

React NativeExpoNode.jsSupabaseGenerative AIDeepSeek

RS Inventory (Microservices)

A scalable hospital inventory system built with Microservices architecture, Node.js, and Docker, featuring a custom API Gateway and rate limiting.

Node.jsExpressMicroservicesDockerMySQLRailwayKubernetes

KangKeliling

A mobile app connecting users with traveling micro-businesses (UMKM) in real-time using Geolocation and Supabase.

React NativeSupabaseGeolocationGoogle Maps APINode.js

Arya's Portfolio