As I joined Coding Dojo, I was tasked to start the data science team in order to provide the teams with meaningful student insights, thus improving a set of core business metrics student retention, student learning experience, revenue recognition, learning science capabilities, marketing and sales efficiency.

My main focus was improving student retention: After a year and half of researching and iterations; I improved student retention by 25% by empowering support managers with an at-risk students identification dashboard, using an explainable, statistical model based on core retention and learning metrics.

In order to tackle business objectives around team efficiency, I helped increase our teams’ daily efficiency by training models and delivering real-time student insights to focus on students who needed the most help.

I had the opportunity to present analyses results to non-technical teams, report and present weekly data findings to C-level executives, shaping the company's data strategy. The strategic thinking included using data to plan Coding Dojo’s adaptive learning strategy based on the LMS data, evaluating marketing channels and sales processes by analyzing their impact on student retention and revenue recognition, etc.

Throughout my work at Coding Dojo, I learned how to manage a team (with one data engineer reporting to me) including regular reviews and goal definition, and I could research ways to improve the Learning Science capabilities & the student bootcamp experience, using for example clustering algorithms for customer segmentation; student survey iterations, A/B testing to improve student data collection; machine learning models with our LMS data: LSTMs, Random Forests... to advance the understanding of our student learning habits.

Managed 1 data engineer, conducted regular reviews & defined goals.

Google Cloud Platform, LMS data, Tableau, Looker, dbt cloud, A/B testing

Deep Learning Classes: dense multi layer perceptrons, classifications, regressions, convolutional Neural Networks, RNNs, LSTMs, Attention layers, etc.

Natural Language ProcessingJavaScript, nools

I joined a semester-long research project aiming at providing tools for low-resourced Ivoirian teachers. I extracted valuable insights from a SQL database storing chatbot conversational metadata to refine interaction design and for usage visualization. Python, numpy, seaborn, sqlite

As a research intern at SEME, I analyzed conversational data and metadata from a chatbot to refine its design. The project, ran jointly with an NGO, aims at helping Ivory Coast primary school teachers fulfill their work aspirations and adopt better teaching methods with support from the community.

Python, seaborn, sqlite, pandas, numpy

I implemented word edit distances for text correction, N-grams and Bayesian text classifiers, Viterbi algorithm for Part-of-Speech tagging. Built a general question answering software in a final group project.

I developed dense, convolutional, recurrent, attention-based neural networks for image recognition and speech-to-text tasks, trained on AWS. Researched and implemented Generative Adversarial Networks for Super Resolution task, achieved a 24+ PSNR and 54+ SSIM on DIV2K dataset.

During the Fall semester, I have been learning about Instructional Design, which is a set of methodologies to develop aligned learning goals, assessment and instruction.

For a Personalized Online Learning class, I designed and implemented digital online cognitive tutors to build adaptive educational courses in Geography. JavaScript, nools

Developed an online educational board game for sustainable shopping learning using iterative design methods and evaluation research. Conducted continuous improvement through user and game testing.

In a project at Carnegie Mellon as part of a two-person team, I designed a ten-hour math course for 2nd graders. Following backward design, we iteratively aligned goals, assessment, and instruction to the educational context, and created an evaluation research module for course improvement.

Using a set of evidence-based 'Big Ideas' related to educational design best practices, I designed the entire course getting insights and supports from state school standards and a kindergarten teacher.

In a two-person team, I designed and implemented an online learning module for South-American geography learning. We tested a new way to learn geography by creating multiple connections among country's representations for learners. We included political, historical and social information, sometimes shared among multiple countries.

From our theoretical approach to fast prototyping, we could apply Cognitive Task Analysis and A/B testing to refine our cognitive model and the module features. The cognitive model used is ACT-R (developed at Carnegie Mellon University), and we personalized instruction by implementing the production rules modeling students' understanding. We used the nools language also developed at Carnegie Mellon University.

Cognitive Tutors, nools (javascript), Cognitive Task Analysis, A/B testing, e-learning design principles

During this 5-month internship in Switzerland at CALYPS SA, I designed and implemented a patients’ hospital stays automatized codification SaaS to improve operational costs, codification efficiency and reliability.

Deep Learning, BERT (camembert), Python, PyTorch, MySQL, Team-based project

I ran an observational study on a 30K+ Junior High School student cohort to identify grade repetition's effects on learning. Statistical Analysis, R

For a two-person semester project, I designed and built a neural network-driven Arduino autonomous car. My role in the project mainly focused on developing and fine-tuning Convolutional NN to drive the car along a curved line, using floor pictures for in-time direction prediction (picture classification). I also helped program the Raspberry Pi 3 using Debian, the Pi was carried on the car together with a camera and an electrical engine with batteries.

Raspberry Pi 3, Arduino, Python, Keras

During a 2-year, four-person group project aiming at improving Companies' Human Resources (detect burn out, better match job seekers and job offers, and others), I co-designed a skill-based job description ontology to match job descriptions to job seekers’ de-identified resumes. I trained and evaluated clustering models and predicted job turnover with classification models (0.98 accuracy) for the same project.

Python, sklearn, seaborn, pandas