A technical management position with dual accountability:1. Led the development of analytical insights, dashboards, and data-driven products to propel the expansion of Gojek's ride-hailing business.2. Nurtured the professional growth of a team comprising two data scientists and two data analysts. Team achievements in H2 2023 includes:- Delivered crucial analyses and dashboards pivotal for two major product launches.- Created a customer segmentation model utilized… Show more A technical management position with dual accountability:1. Led the development of analytical insights, dashboards, and data-driven products to propel the expansion of Gojek's ride-hailing business.2. Nurtured the professional growth of a team comprising two data scientists and two data analysts. Team achievements in H2 2023 includes:- Delivered crucial analyses and dashboards pivotal for two major product launches.- Created a customer segmentation model utilized across the organization for enhanced analyses and modeling.- Assessed statistically significant impacts across multiple product launches via controlled A/B tests and synthetic control methodologies.- Implemented an address ranking service, resulting in a 11% increase in click-through rate. Show less

Held a senior individual contributor (IC) role responsible for end-to-end data science solutions, ranging from the scoping of business challenges into data-driven opportunities to the deployment of models into production. Key project highlights include:- Smart reply. Developed NLP models that enable quick-tap for replying to a message, saving the user the hassle of typing. The NLP model has evolved from a simple keyword detection, then a BERT-based classifier, to a Seq2Seq-based… Show more Held a senior individual contributor (IC) role responsible for end-to-end data science solutions, ranging from the scoping of business challenges into data-driven opportunities to the deployment of models into production. Key project highlights include:- Smart reply. Developed NLP models that enable quick-tap for replying to a message, saving the user the hassle of typing. The NLP model has evolved from a simple keyword detection, then a BERT-based classifier, to a Seq2Seq-based generative model. Adoption rate jumps to 50% from 20% after the adoption of the model. The PyTorch model is still running in production, serving more than hundreds RPS at peak time. Link to blog: https://www.gojek.io/blog/working-in-the-central-analytics-and-science-team.- OmniBERT. Trained BERT language models on Gojek's internal corpus of 200 million sentences. This LM is now powering all BERT-based text classifiers in production, where downstream evaluation shows a 20% increase in text classification accuracy on the internal Gojek's corpus.- Promo card recommendation. Built a reinforcement learning-based ranking system for promotion cards on the Gojek app using the Epsilon-Greedy and Thompson sampling policies. Experiment result shows a 20% increase in the click-through rate and 1% increase in the completed order. Show less

Headed the data science and engineering team of Hukumonline, driving the creation of a strong data infrastructure from the ground up while growing a team consisting of two data scientists and two data engineers. Noteworthy projects include:- Legal document parser. Developed a parser essential for converting legal text into structured data, vital for advanced text analysis. By redefining the parsing problem as a sequence labeling task with CRF, we achieved a significant upgrade from the… Show more Headed the data science and engineering team of Hukumonline, driving the creation of a strong data infrastructure from the ground up while growing a team consisting of two data scientists and two data engineers. Noteworthy projects include:- Legal document parser. Developed a parser essential for converting legal text into structured data, vital for advanced text analysis. By redefining the parsing problem as a sequence labeling task with CRF, we achieved a significant upgrade from the previous CFG parser, boasting an impressive 97% F1 score on properly formatted documents.- Data pipeline. Engineered a dependable ETL pipeline leveraging Airflow, Spark, and proprietary Python libraries, which remains operational in production. This pipeline drives critical company functions such as regulation indexing, data warehousing, feature extractions, and MLOps, ensuring seamless data management and operational efficiency.- Business intelligence. Utilized our BI dashboards and reports to drive strategic initiatives, resulting in impactful company-wide decisions such as optimizing newsletter content, refining subscription tiers, and enforcing maximum news article length. Show less

An applied NLP research role whose responsibility is to research and prototype NLP approaches to enhance Kata.ai's product and turn them production. Project highlights:- Named Entity Recognition (NER). Addressed Kata.ai's challenge of limited annotated data by devising a solution that involved training a Bi-LSTM CRF model on a standard corpus and leveraging pre-trained weights. Implemented in Keras and Tensorflow, the model achieved an average F1 score over 80% in production.-… Show more An applied NLP research role whose responsibility is to research and prototype NLP approaches to enhance Kata.ai's product and turn them production. Project highlights:- Named Entity Recognition (NER). Addressed Kata.ai's challenge of limited annotated data by devising a solution that involved training a Bi-LSTM CRF model on a standard corpus and leveraging pre-trained weights. Implemented in Keras and Tensorflow, the model achieved an average F1 score over 80% in production.- Automated Speech Recognition (ASR). Pioneered the development of an ASR prototype tailored for the Indonesian language using deep learning ASR models such as DeepSpeech and Wav2Letter.- Publication. Multi-Task Active Learning for Neural Semantic Role Labeling on Low Resource Conversational Corpus. ACL 2018 Workshop on Deep Learning Approaches for Low-Resource NLP. Link to publication: https://aclweb.org/anthology/W18-3406. Show less

Implemented Ericsson solutions within clients' networks by loading configurations onto new hardwareto optimise the network capacity, throughput, and efficiency. Key project includes the deployment of the biggest fixed IP Multimedia Subsystem (IMS) network in the Asia Pacific with Telkom Indonesia. Activities include:▪ support presales activities, including pre-studies▪ analyse the customers’ existing systems, needs, and requirements▪ establish test specifications and procedures… Show more Implemented Ericsson solutions within clients' networks by loading configurations onto new hardwareto optimise the network capacity, throughput, and efficiency. Key project includes the deployment of the biggest fixed IP Multimedia Subsystem (IMS) network in the Asia Pacific with Telkom Indonesia. Activities include:▪ support presales activities, including pre-studies▪ analyse the customers’ existing systems, needs, and requirements▪ establish test specifications and procedures for the IMS solution developed▪ provide deployment, installation, configuration, integration, testing, and support services of IMS product and technology solutions▪ document the implemented solution and give training to customer about the solution Show less