Arjaan Buijk

I developed a data pipeline and a machine learning model to predict university student application probability based on time-stamped engagements.

The data pipeline extracted millions of records from several SQL databases and generated features for the machine learning model. The end result of the data-pipeline was a pandas DataFrame written to an S3 bucket.

The machine learning pipeline loaded the pandas DataFrame and trained a custom TensorFlow model used by the university admissions department to identify the most promising prospective applicants.

I designed and implemented NLU data, dialog stories, forms, and custom actions (Python) for a Rasa financial chatbot starter-pack.

The implementation allowed the user to switch conversation contexts and receive guidance to succesfully complete conversations.

I created and taught a Rasa advanced deployment workshop, which is hosted on Udemy.

The workshop teaches how to deploy a Rasa assistant on Kubernetes in the Google Cloud Platform (GCP) and how to use a CI/CD pipeline with GitHub Actions.

I developed a chatbot with Rasa and Elasticsearch and deployed it to a single node Kubernetes cluster on AWS.

The chatbot provides an alternative interface to a web-based expert system.

The data pipeline creates word and sentence embeddings from web-scraped data and injects them into an Elasticsearch database. The embeddings are created using pre-trained machine learning models from TensorFlow Hub.

The chatbot listens to questions from users and finds the most similar result by querying Elasticsearch.

An industrially proven software package for the computer simulation of industrial forging processes. It combined a familiar and intuitive Windows graphical user interface with a robust solution procedure to provide unprecedented accuracy and speed in forging simulations.

I was both a solver and a front-end developer.

This portfolio project demonstrates a Flask-based implementation of a secure user model with registration, email confirmation, reset, and login, combined with an example of a catalog of items grouped into categories.

Registered users can CRUD the catalog via a web front end or via a REST API that adheres strictly to the JSON API 1.0 specification. The REST API supports advanced searching and filtering. End-to-end scenarios are demonstrated via a Python client written in a Jupyter notebook.

I developed a convolutional neural network regression in Python using Keras with a TensorFlow back end to drive a car virtually around a track. I improved on the well-known NVIDIA neural network for end-to-end deep learning for self-driving cars and achieved a 100% success rate on the most difficult track.