Projects
Research prototypes, hackathon winners, and deployed products, sorted by recency. Click any tile to read more.
PM2.5 air quality forecasting system using Random Forest Regressor, deployed as a full-stack web app for BMKG Yogyakarta alongside rainfall forecasting from climatology station data.
Lightweight Bahasa Indonesia text-to-speech system based on fine-tuned Piper/VITS models, with male and female voices optimized for CPU-only inference.
Voice assistant for lab automation at DTETI UGM. Whisper and a fine-tuned Piper voice run locally on a Raspberry Pi, with LLM command understanding and N8N workflows for lights, HVAC, and projectors.
Android app for real-time BISINDO and Indonesian translation. It runs fully on-device and was built for Google Gemma 4 Good Hackathon.
Agentic RAG system live at ask.pradanayahya.com. It combines hybrid retrieval, citations, and agent workflows for document-grounded question answering.
2nd Place at DSC LOGIKA UI 2025. Semi-supervised classification pipeline for Indonesian traditional house images using DINOv3 features and pseudo-labeling; 97.21% test accuracy.
3rd Place at ACTION UNESA 2025 Data Mining Competition. Named Entity Recognition pipeline for Indonesian culinary text using IndoBERT; Micro-F1 0.8411 on the Kaggle leaderboard.
Sentiment analysis and topic modeling on public response to Prabowo’s decision granting abolition to Tom and amnesty to Hasto.
Sentiment analysis and topic modeling on the Tom Lembong verdict. BERT for sentiment classification and BERTopic for topic modeling.
Fine-tuned sentiment-based RoBERTa for hate speech classification on Indonesian text.
Fine-tuned IndoBERT base for sarcasm classification on Indonesian social-media text.
Champion at DATATHON 2025. Multimodal retail analytics platform turning ordinary CCTV footage into customer-behavior signals and strategy.
Movie recommendations based on your mood. Built with semantic search, sentiment analysis, and mood-based filtering.
RAG-based assistant that uses ModernBERT for semantic retrieval. Upload your PDFs and get precise, citation-backed answers.