Christodoulos Constantinides

Introduction

I am a data scientist at IBM in New York where I build timeseries models for predictive maintenance across different industries.
I also do research on how we can domain-adapt LLMs for the industrial domain.
I have an MS in Data Science from Columbia University and a BS in Computer Science from University of Cyprus.

Some topics that I am interested in:

Few-shot learning for Timeseries Classification. Motivation: Predictive Maintenance where there are very few sensor data examples from engine failures available
Timeseries Anomaly Detection. Motivation: Plenty of healthy engine sensor data and rare events of engine failures (anomalies)
LLM Domain adaptation. Motivation: Domain-adapt LLMs for the industrial domain
LLM Knowledge Distillation. Motivation: Distill knowledge from larger models into smaller ones that have smaller inference costs

Education

Columbia University MS in Data Science	Jan 2021 – May 2022 New York, NY
Selected Courses: Machine Learning, Applied Deep Learning, Statistical Inference & Modelling
Thesis: Quantifying the impact of Climate Change on Socially Vulnerable Population Built a geospatial dataset for climate justice research with social vulnerability, climate disaster and housing data Handled missing values, removed correlation and multicollinearity Predicted the number of evictions in a given area using XGBoost model with Poisson Loss Interpreted the predictions and the impact of each feature using SHAP
Selected Courses: Data Mining, Algorithms & Data Structures, Probability & Statistics, Linear Algebra, Calculus
Thesis: Intelligent User Authentication Systems Developed a graphical password security analysis web application in Django to detect objects in an input image and perform informed brute-force attack to estimate the number of guesses needed to crack a picture password Integrated computer vision algorithms on TensorFlow and Amazon Rekognition

Columbia University
MS in Data Science

Jan 2021 – May 2022
New York, NY

Selected Courses: Machine Learning, Applied Deep Learning, Statistical Inference & Modelling

Thesis: Quantifying the impact of Climate Change on Socially Vulnerable Population

Built a geospatial dataset for climate justice research with social vulnerability, climate disaster and housing data
Handled missing values, removed correlation and multicollinearity
Predicted the number of evictions in a given area using XGBoost model with Poisson Loss
Interpreted the predictions and the impact of each feature using SHAP

University of Cyprus
BS in Computer Science

Jan 2017 – Jan 2021
Nicosia, CY

Selected Courses: Data Mining, Algorithms & Data Structures, Probability & Statistics, Linear Algebra, Calculus

Thesis: Intelligent User Authentication Systems

Developed a graphical password security analysis web application in Django to detect objects in an input image and perform informed brute-force attack to estimate the number of guesses needed to crack a picture password
Integrated computer vision algorithms on TensorFlow and Amazon Rekognition

Experience

IBM Data Scientist	Jul 2022 – Current New York, NY
Predictive Maintenance on critical Industrial Engines for a Big Tech Data Center using Maximo to predict Failures, Anomalous Behaviors and Survival Analysis and increase their reliability Modelling the health of the train tracks for a national railway network to do Condition-Based Maintenance Data cleaning, analysis, modelling and deployment of live end-to-end models for near-realtime predictions on high frequency IoT sensor data Forecasting energy consumption in Energy Plants using Timeseries Foundation Models Predicting Cancer-Associated Venous Thromboembolism to aid the doctor’s treatment decision using Deep Survival models implemented in PyTorch Applied research on fine-tuning transformer-based models for domain adaptation and improving word embeddings
Optimized the Click Through Rate (CTR) of the on-site articles using reinforcement learning to learn the optimal recommendation strategy Used Multi-Armed Bandit for learning the optimal recommendation strategy for the different times of the day Distributed processing of the event stream using Spark and storage of the end result in PostgreSQL to reduce processing time
Deployed models for continuous face and voice authentication with distributed work queues using RabbitMQ Built a cross platform mobile application using Flutter for alternative two factor authentication methods Developed the authentication server in Django and the admin dashboard with user analytics and shipped with Docker

IBM
Data Scientist

Jul 2022 – Current
New York, NY

Predictive Maintenance on critical Industrial Engines for a Big Tech Data Center using Maximo to predict Failures, Anomalous Behaviors and Survival Analysis and increase their reliability
Modelling the health of the train tracks for a national railway network to do Condition-Based Maintenance
Data cleaning, analysis, modelling and deployment of live end-to-end models for near-realtime predictions on high frequency IoT sensor data
Forecasting energy consumption in Energy Plants using Timeseries Foundation Models
Predicting Cancer-Associated Venous Thromboembolism to aid the doctor’s treatment decision using Deep Survival models implemented in PyTorch
Applied research on fine-tuning transformer-based models for domain adaptation and improving word embeddings

Condé Nast
Machine Learning Engineer Intern

Jun 2021 – Sep 2021
New York, NY

Optimized the Click Through Rate (CTR) of the on-site articles using reinforcement learning to learn the optimal recommendation strategy
Used Multi-Armed Bandit for learning the optimal recommendation strategy for the different times of the day
Distributed processing of the event stream using Spark and storage of the end result in PostgreSQL to reduce processing time

Cognitive UX Ltd.
Software Engineer

Mar 2020 – Jan 2021
Nicosia, CY

Deployed models for continuous face and voice authentication with distributed work queues using RabbitMQ
Built a cross platform mobile application using Flutter for alternative two factor authentication methods
Developed the authentication server in Django and the admin dashboard with user analytics and shipped with Docker

Projects

The Effects of Plasticity Functions on Neural Assemblies	Fall 2021 New York, NY
Modified the Assembly Calculus simulation code to implement more biologically plausible plasticity functions (Oja, STDP) Analyzed the effects of the different plasticity functions to the resulting neural assembly
Built the authentication component as a RESTful service in Django and documented with Swagger Developed automated unit tests and stress-tests Collaborated with other organizations for integration and testing and conducted case study with end-users for usability testing

The Effects of Plasticity Functions on Neural Assemblies

Fall 2021
New York, NY

Modified the Assembly Calculus simulation code to implement more biologically plausible plasticity functions (Oja, STDP)
Analyzed the effects of the different plasticity functions to the resulting neural assembly

SERUMS H2020 - University of Cyprus

Jun 2019 – Sep 2019
Nicosia, CY

Built the authentication component as a RESTful service in Django and documented with Swagger
Developed automated unit tests and stress-tests
Collaborated with other organizations for integration and testing and conducted case study with end-users for usability testing

Technical Skills

Languages: Python, Java, R, C, C++, PHP, JavaScript, Dart, ARMv8, Databases: MySQL, Neo4j, MongoDB, Cloud Computing: Spark, Databricks, Hadoop, Docker, Machine Learning / Data Science: PyTorch, TensorFlow, Hugging Face, Numpy, Pandas, Web Development: Django, HTML, CSS, JavaScript, Parallel Processing: Multi-Threaded Programming, CUDA

Awards

Outstanding Technical Achievement Award - IBM

Apr 2023

Cypriot ICPC competitive programming contest - 2nd and 3rd place

Sep 2018, Jun 2019