Carina van der Zee

Designed and developed a web-based application that automates social media marketing efforts using generative AI. I worked with Next.js to create an intuitive, user-friendly front-end interface, ensuring a seamless user experience. Also, I developed a robust back end using Python and FastAPI and leveraged APIs from Twitter, Instagram, and LinkedIn to enable the app to interact with these platforms and automate social media marketing efforts. Finally, I incorporated the OpenAI API into the tool to utilize the power of GPT-4 for content generation, enhancing the creativity and diversity of marketing content, integrated Stripe for handling payments, and deployed the back end on AWS.

Created and helped implement a detailed AI strategic roadmap for several startups. My role encompassed providing strategic direction, technical expertise, and project management to drive the AI transformation journey of startups.

I conducted a thorough analysis of the startup's current systems and processes, identifying gaps and determining areas where AI could provide the most significant value. Using the startup's specific needs and goals, I provided recommendations for adopting the most suitable tech stacks and AI models. Given the critical nature of resource allocation in startups, I estimated the required person hours and the specific skill sets needed for the project's success.

Developed algorithms to segment clients for a trading desk. The project aimed to classify clients based on unique trading behavior characteristics, enabling the bank to offer more personalized service and improve operational efficiency.

I compared a range of clustering methods, including k-means, hierarchical, and DBSCAN, among others, to ensure the most effective algorithm was selected for the segmentation task. I designed summary statistics based on high-frequency market data and higher-level client/economic data to feed into the clustering process, optimizing the accuracy and efficiency of client segmentation, and explored several dimension reduction techniques, including PCA and auto-encoders, to handle the vast amount of relevant data available. Finally, I collaborated with the business to define KPIs to monitor and measure the project's success, ensuring alignment with the bank's strategic objectives.

To wrap up the project, I wrote technical documentation and implemented the tool in production in Python. I collaborated closely with the stakeholders at the trading desk and the rest of the business.

Assisted a MedTech startup in designing a robust and effective clinical trial protocol, ensuring scientific validity and practical feasibility.

I calculated the necessary sample size required for the trials to provide statistically significant results, guaranteeing the trials' reliability. To understand potential outcomes and risks associated with the trial design, I ran simulations under different setups and assumptions. I created visualizations of study designs and inherent uncertainty, providing an easy-to-understand view of potential trial outcomes. Finally, I considered several statistical tests, including the t-test, analysis of variance (ANOVA), chi-squared test, Fisher's exact test, and Wilcoxon rank-sum test for the setup of the medical experiments.

Developed and backtested fixed-income trade strategies around market events based on several predictive models for market indicators. Several techniques were used, including time series modeling, copula techniques, deep learning, and reinforcement learning.

A Python-based statistical model that predicts mortgage prepayment rates given the portfolio characteristics. I conducted the complete data analysis, feature engineering, model development, and production code implementation.

Trained, tested, and compared several predictive models (VARMA(X), XGBoosted tree, RNN, copula) to predict the price moves of multiple cryptocurrencies. Compared several predictive power for different time windows and investigated potential explanatory variables (besides AR/MA components).

Research into methods and algorithms for clustering high-frequency time series data.

I retrieved the data from KDB, a high-frequency market database. Also, I investigated multiple approaches to time series clustering, such as dimensionality reduction combined with traditional clustering methods, dynamic time warping, and image-based clustering using a convolutional neural network (CNN). I executed all aspects of the project using Python.

Upon completion, I presented and delivered a detailed writeup of the project results to key stakeholders.