Open to opportunities

Marsa Thoriq

@marsathoriq

Message

Machine Learning Engineer specializing in real-time fraud and risk detection for e-commerce.

Indonesia

Message

What I'm looking for

I want to build and deploy large-scale real-time fraud/risk ML systems end to end—partnering with ops and product, iterating quickly on signals and models, and owning monitoring plus incident response.

I’m a Machine Learning Engineer / Data Scientist focused on large-scale fraud and risk detection for e-commerce. I own the full cycle—from pattern discovery and model training to real-time deployment, monitoring, and incident response—while partnering closely with business and ops teams to translate real fraud behavior into actionable detection strategies.

In my recent work, I’ve engineered 30+ real-time fraud strategies inside a fraud engine, using statistical anomaly methods, device/address signals, and LLM-assisted techniques to reduce subsidy loss and blocking of fraudulent orders. Previously, at Tokopedia, I designed anomaly detection and seller risk systems (including graph-based risk with Neo4j and credit scoring models) that improved precision, coverage, and revenue-impacting segmentation, backed by multiple national data science competition wins.

Experience

Work history, roles, and key accomplishments

Machine Learning Engineer

ByteDance (TikTok Shop)

Dec 2024 - May 2026 (1 year 5 months)

Engineered and deployed 30+ real-time fraud strategies in the fraud engine, addressing incentive abuse and address codeword fraud with PySpark pipelines, LLM-based NER, and model consensus reviewing. Owned monitoring and incident response, including a checkout SLA-bypass exploit mitigation and address clustering for coordinated fraud using transformer embeddings and Faiss.

Real Time Fraud Address Codeword Fraud Pyspark Few Shot Learning Consensus LLM Review CatBoost Embeddings Faiss GPU Index K NN Graph Based Connection Limits Dynamic Blacklists

Data Scientist

Tokopedia

Jul 2022 - Dec 2024 (2 years 5 months)

Designed anomaly detection and seller fraud risk systems for promo abuse, cash advance fraud, and coordinated seller networks. Built credit scoring and propensity models (LightGBM/Optuna) and a fraud graph approach using Neo4j weakly connected components.

Anomaly Detection Cash Advance Fraud Detection CatBoost Seller Risk Modeling Feature Engineering Neo4j Weakly Connected Components LightGBM Optuna Credit Scoring A B Testing Model Evaluation

AI Engineer

Pinhome

Jan 2021 - Jul 2022 (1 year 6 months)

Built a property listings web-scraping pipeline to expand and enrich the company dataset. Developed XGBoost price prediction and EfficientNet-based similar listing detection with Optuna and Airflow for MLOps.