Research & Publications

A multi-agent reinforcement learning system for portfolio management on the Indonesia Stock Exchange, trained and evaluated inside a market simulator that models real IDX trading rules. Developed from my undergraduate thesis.

• Universe of 53 Indonesian stocks on 5-minute data.
• Five SAC agents blended through a PPO meta-controller.
• Simulator of IDX microstructure: ARA/ARB price limits, asymmetric transaction costs, and lot sizes.
• Ablations showed the gains come from policy diversification rather than learned regime adaptation.

+7.23%

Return over 11 walk-forward windows, vs LQ45 at −12.89%

−19.11%

Loss on a bear-market holdout, vs LQ45 at −23.22%

Computer vision models that detect and classify road damage from satellite and aerial imagery, supporting post-disaster road network restoration planning.

• Detection and classification of road damage from satellite and aerial imagery.
• Built to support restoration planning of road networks after disasters.
• Cross-disciplinary collaboration with civil engineering researchers at ITB and ITERA.

001514719

Computer-program copyright registration number

ITB × ITERA

Joint research with civil engineering