Resume

Stealth AI Startup · Kuala Lumpur, Malaysia

• Fine-tuned cross-domain Speech & Text models with LLM-RAG integration for Voice Assistants, substantially reducing error rates against open-source baselines for APAC and MENA languages
• Engineered and deployed high-throughput, low-latency audio pipeline modules (Transcription, LID, VAD, Emotion Detection, Audio Spectrogram Transformers), optimizing inference for scalable production Voice Agent frameworks
• Overhauled Active Learning pipelines using LLM-assisted auto-labeling and human-in-the-loop verification, scaling up dataset coverage and annotation accuracy
• Built end-to-end data and training automation loops, leveraging DVC and GitHub Actions for event-triggered CI/CD orchestration and rapid, reproducible ML experimentation

Stealth Startup · Remote

• Detected and tackled fraudulent activity, reducing customer complaints and averting recurring monthly losses
• Reviewed fraud networks from past cases using Graph DS and historical activity, including churn on complaints
• Explored LLM and agentic workflows for internal fraud-detection tooling
• Built data-driven alerting for risk prevention and mitigation (in-app loopholes, catastrophic alerting)

TikTok · Jakarta, Indonesia

• Harmonized cross-platform search algorithms and system architectures between Tokopedia and TikTok Shop post-acquisition, fine-tuning retrieval models with local marketplace nuances to lift search relevance
• Spearheaded core phases of TikTok's end-to-end ML lifecycle, designing and scaling pipelines for data streaming, automated model training, search indexing, and high-dimensional Vector DB infrastructure

Tokopedia · Jakarta, Indonesia

• Architected Query Understanding algorithms for Search Personalization, lifting search relevance, conversion, gross transaction value, and monthly ads revenue
• Engineered a multi-label Query-to-Product Type assortment system for IR pre-filtering, driving gains in conversions, transaction value, and monthly ads revenue
• Scaled deep learning and LLM solutions in production (real-time classifiers, LLM automated labelers), leveraging model distillation and quantization to cut latency and compute cost while holding baseline accuracy

Pashouses · Jakarta, Indonesia

• Engineered a multi-modal Automated Valuation Model (AVM) for regional housing prices using tabular, text, and image data, keeping the error rate low while greatly accelerating processing throughput
• Designed end-to-end data engineering and ETL pipelines for property analytics, cutting manual data ingestion effort while keeping the system reliable

Bandung Institute of Technology (ITB)

• Thesis: Computational Fluid Dynamics for Wave Propagation using the Shallow Water Equation
• Focus area in Applied Mathematics (Optimization, PDE, Financial Mathematics)
• Core math coursework: Linear Algebra, Real & Complex Analysis, Multivariable Calculus
• Intermediate statistics: Probability Theory, Mathematical Statistics
• Activities: International Mathematical Modeling Competition (fish population dynamics); Quantitative Finance portfolio-construction challenge

Bringing modern NLP to Indonesian & its local languages

• LLMs underperform on low-resource languages due to scarce data
• Many Indonesian local-language resources remain untapped by modern NLP
• Contributing to projects (with talented collaborators across Indonesia) that address this gap

Expanding generic NLP toolkits

• Widely-used NLP toolkits often lack Indonesian and local-language support
• Aiming to help democratize tooling and knowledge across languages

Low-resource NLP research

• Several Indonesian local languages are syntactically transferable from Indonesian without major semantic degradation
• Dedicated to bridging NLP into low-resource languages via this approach
• Aspiring to pursue a higher degree in an environment that enables driving this vision