An engineer who turns messy government data problems into AI systems that run in production. He cut Ministry of Finance budget analysis from 5 days to 20 minutes, and built a legal research platform that indexes 297 thousand regulations from 37 government sources.
His focus is system architecture, large-scale data, and language models for legal documents.
Problem: APBN and APBD account schemes are incompatible, so cross-system public spending analysis is slow and expensive. Result: reconciled 1.83 million budget records (about USD 224.4 billion of public spending), cut analysis time from 5 days to 20 minutes, cut cost per analysis from about USD 28 thousand to about USD 50, and deployed on AWS Region Jakarta to meet data residency rules.
Problem: Indonesian law is spread across 37 government sites and hard to trace. Result: 297 thousand regulations acquired and indexed into about 3.48 million vectors, 2.57 million citation relations mapped at about 99% precision, self-hosted OCR on a 16-GPU RTX 5090 cluster saving about 92% of OCR cost, and serving about 100 daily active users in production.
