Anthony Khong

One of Indonesia's retail giants wanted to bridge the gap between executives in the HQ and their clients. They wanted to be reminded and consider their typical clients' needs to inform them about high-level strategic decisions.

I led a research team that used multiple years of transactions to come up with about a dozen customer archetypes that are easy to relate to while still backed by data. The final persona deliverables were later used to inform executive decisions on marketing campaigns, brand positioning, and product offering gaps.

A large-scale constrained optimization application to optimize production planning for a manufacturing company in Indonesia. The app considers over a hundred thousand constraints related to production quality, stock availability, and stock replenishment possibilities. The project involved creating a mathematical model to represent the production processes, developing the application using mainly Python, OR-Tools, and Django, and deploying the application on AWS.

I worked with one of Indonesia's retail giants to develop a production planning application, which involves an optimal stock replenishment algorithm. The algorithm is comprised of a supervised learning model that uses several years of past transaction data to predict the sales schedule of a particular stock keeping unit (SKU). It takes into account features such as seasonality, trends, special dates, and item characteristics. The predictions are used to guide merchandise planners in making the right decisions on the timing and quantity for stock replenishments.

I am the author of Geni, a Clojure DataFrame library that runs on Apache Spark. The main idea of Geni is to provide an idiomatic DataFrame API in Clojure while leveraging on the already mature Spark ecosystem. As a result, Geni is stable, performant, and has all the important Spark features, including Spark ML.

I started the project, designed the API, wrote most of the documentation, and provided examples, notably one for deploying Geni to Google Cloud's Dataproc. The project has 100% test coverage. It is installable with a few Bash lines and includes a cookbook to optimize the getting started experience. It has been used as the main data analysis and machine learning library for a couple of past projects.