I'm an applied scientist working on AI safety. I want models to surface what they actually know, to hold back when they should, and to do both reliably for the languages and contexts they were never trained on. Concretely, that means trustworthy multimodal systems that are reliable when it matters and relevant to the places they operate in.
At Monash University, I'm working with Associate Professor Risqi Saputra and Professor Taufiq Asyhari on conformal prediction for vision-language navigation. The goal is simple to state and hard to achieve. When a model claims confidence, we want that confidence backed by provable bounds.
The other thread in my work is cultural inclusion. Southeast Asia is one of the world's most linguistically and visually diverse regions, and it's almost entirely absent from the training data and benchmarks that define what modern AI can and cannot do. I've been building the datasets, benchmarks, and adaptation methods to change that, with SEACrowd and through my own research.
This Month · updated June 2026
- Writing up my M.Sc. thesis on conformal prediction for vision-language navigation, sharpening the abstention and set-efficiency results
- Following up the multilingual VLM study: turning cross-lingual activation steering for abstention from a hackathon proof into a real method
- Testing whether abstention signals and honesty directions transfer across languages, or fail silently outside English
Tech Stack
Research Agenda
I work on one question from two sides. Getting a model to surface what it actually knows, and to hold back when it should, reliably, for the languages and contexts it was never trained on. Three open problems I would bet on, and the place I pressure-test them. Each sits in the gap between two communities that rarely talk to each other, which is usually where the tractable and neglected work hides.
Saying what you know, in any language
Models often state a confident falsehood while their own representations encode the correct answer, and the gap widens when the modalities disagree or the query is in a low-resource language. I want systems that defer to their own better-supported internal belief, and I want us to actually measure how this failure shifts across languages, because almost no one does. A model that confidently states a falsehood it could have flagged is a deployment hazard, not a rounding error.
Whether our safety tools even work outside English
Probing and activation steering are becoming the core tools for reading a model's internal state for deception, refusal, or harmful intent. Almost all of that work is validated in English, and we do not know whether a probe or a steering vector transfers to another language or fails silently outside it. If internal monitoring holds for English and breaks for the languages most of the world actually uses, the safety stack carries a language-shaped hole and a false sense of security with it.
Making "I do not know" trustworthy, with guarantees
Deployed and agentic multimodal systems need to know when not to act, but their confidence signals are poorly calibrated and carry no guarantees. Conformal prediction gives distribution-free coverage guarantees, and extending it to multimodal, sequential, and free-form generation is largely open. An agent that acts confidently when it should defer is a direct path to loss of control. Calibrated, guaranteed abstention is one of the few safety levers I can make rigorous.
Where it gets tested
These ideas have to survive contact with messy data. I pressure-test them in vision-language navigation, the subject of my thesis, in multimodal models for earth observation published in IEEE and Remote Sensing of Environment, and in open multilingual AI for Southeast Asia with SEACrowd and SEA-VL. The regions and sensors that English-first, benchmark-first AI never really sees.
Track Record
WORK EXPERIENCE
-
OCT 2024 – PRESENT
SEACrowd - Researcher, Multimodal & Vision-Language
Open-science research collective · seacrowd.github.io
-
FEB 2025 – NOV 2025
Artefact - Senior Data Scientist
French AI consulting · Founding technical member, Jakarta office
-
DEC 2022 – JAN 2025
Monash University - Research Associate
Global research consortium: Monash, UQ, UCL, Nottingham
-
JUN 2021 – JUN 2023
GDP Labs (GLAIR.ai) - Senior Data Scientist / ML Engineer
AI consulting, backed by a major Indonesian conglomerate
-
JAN 2021 – JUN 2021
Jakarta Smart City - Data Scientist
Indonesia's smart city government initiative
PATENT
-
ISSUED JUNE 2025
Fish & Shrimp Pond Detection via Satellite Imagery
IDS000010594
TALKS
-
MAY 2026
PyPalu, Python for Localized Context
Python community meetup · Sulawesi Tengah, Indonesia
-
2025
MUSE, How Data Science Differs in Each Sector
Panel speaker · Monash University Indonesia
-
OCT 2024
Bank of Indonesia, Data Synthesis, Privacy & Responsible Data Management
Invited talk · Bank of Indonesia
-
Q3 2026 OPEN
Available for conference talks, podcasts, and panel invitations
Topics: trustworthy AI, conformal prediction, AI for Southeast Asia
EDUCATION
-
EXPECTED SEPT 2026
Monash University
Master of Data Science · GPA: 4.0/4.0
-
JUN 2026
BlueDot Impact, Technical AI Safety
Cohort intensive · alignment, interpretability, red-teaming, AI control
-
DECEMBER 2019
Monash University
Bachelor of Computer Science
-
MAR – DEC 2019
Monash CURIE Compass
Mentee · Centre for Undergraduate Research Initiatives and Excellence
-
JUN – DEC 2018
Nanyang Technological University
Computer Science Exchange Programme · Singapore
-
2019
Udacity, Deep Learning Nanodegree
Neural networks, CNNs, RNNs, GANs · Facebook AI scholarship recipient
TEACHING
-
MAY 2026
Monash University PGIE, Industry Judge
Faculty of IT Postgraduate Industry Experience
-
MAY 2026 – JUN 2026
IBM SkillsBuild
Capstone Project Advisor
-
MAY 2026 – JUN 2026
Coding Camp powered by DBS Foundation
Capstone Project Advisor
-
JAN 2024 – JAN 2025
Bangkit Academy (Google, Gojek, Traveloka)
ML Instructor & Capstone Advisor
-
NOV 2024
Bina Nusantara University
Guest Lecturer, Computer Vision
Recognition
- CHAMPION Microsoft Azure Virtual Hackathon APAC
- TOP 3 CamvsCovid, University of Cambridge
- BEST COMMUNITY TRACK Cal Hacks 8.0, UC Berkeley
- TOP 10 Reboot the Earth, United Nations Technology Innovation Labs
- MOST VISIONARY RESEARCH Mining Spatial Data Intelligence Research Hub, IISF 2024
- SCHOLARSHIP Monash Indonesia Inaugural Welcome Scholarship
- SCHOLARSHIP Deep Learning Nanodegree, Facebook & Udacity
- SCHOLARSHIP Jeffrey Cheah Entrance Scholarship
- TOP 3 Monash Hackathon
- SCHOLARSHIP Monash Travel Grant
- HONORABLE MENTION Monash University Coding Competition
PROFESSIONAL SERVICE
- Industry Judge: Monash FIT PGIE 2026, evaluated 6 cross-disciplinary teams (36 students) from Master of AI, Data Science, and IT presenting real-world industry projects.
- Peer Reviewer: IEEE IGARSS 2026, Premier global remote sensing and geoscience symposium.
- Technical Judge: Cal Hacks 8.0, CruzHacks 2022, iNTUition v8.0, evaluated 50+ projects across three international hackathons.
- Open Source Contributor: SEACrowd collective, building AI infrastructure and data democratization for Southeast Asia.
