Anne Charlotte Leysen

A fintech startup with an AI solution. I developed the back-end data pipeline from scratch, pulling in company filing data, cleaning, joining, standardising, and storing the data. I automated the pipelines using Airflow on AWS and developed an API for the front-end app to access the prepped data.

I also integrated an AI assistant endpoint into the API for the front end to ask questions about the company data. This was achieved with retrieval augmented generation (RAG) by indexing and vectorizing the input data.

I worked with the founder of this healthcare business to interpret the data collected from EDI files in the most insightful way. I developed numerous dashboards for the founder to present to her clients that provided actionable insights to help care teams administer a better service for their clients (patients) in completing insurance claims.

I worked on Snowflake to query the data via SQL. I pulled this data into Google Sheets to create intractable charts. I then translated these charts into Hex App, a more integrated visualization tool that can handle larger volumes of data than Google Sheets. The Hex tool also incorporated Python scripts from a more complex level of analysis.

These visualizations were automatically refreshed every day to meet the stakeholder needs.

I also worked closely with the founder weekly to ensure she received the highest quality insights for her clients.

I created a home hazard model to predict the likelihood of hazards in homes across the UK, including dampness and mold, excess cold, fire, falls, and electrical.

The training dataset included address-level data on the buildings' construction, e.g., wall material, insulation materials, boiler type, etc. This was enriched with other data, such as weather and poverty metrics.

Data cleaning techniques needed to be heavily deployed for address cleaning to ensure better matching across different data sets. Automated cleaning scripts were created to ensure an accurate, model-ready dataset to work with.

I then used supervised learning models such as logistic regression, SVM, and random forests to train the data. and experimented with clustering algorithms and Naive Bayes classifiers.

The final outputs included low, medium, and high-level risk predictions for thousands of households across the UK for each type of hazard. These results were delivered to councils across the UK to streamline their proactive home hazard prevention strategies.

As part of a mathematical consultancy team, I created a 2D Python model to simulate large crowd movements within public areas during a crisis situation, e.g., an attack or crushing incident. The model simulated where each individual would most likely move every second during a 10-minute period. This included what kind of decisions would be made per individual; for example, would they walk, run, hide, or freeze from danger?

Detailed UK OS Maps were ingested to create a base landscape of buildings, streets, and pavements for the crowd to interact with. Individuals had different characteristics based on their demographic split.

The entire codebase was created with Python and hosted in AWS cloud. A UI/front end was built for users to create new simulations and view simulation results.

I also participated in in-depth, real-world research on crowd behaviors and psychology for each new feature. I was tasked with translating the research into a form that could be coded, and I helped code many critical features implemented in the model.

I was the first data hire in this financial research company. I developed a data platform and automated pipeline that pulled data via HubSpot, WooCommerce, and Google Analytics APIs. This data was pulled daily into a cloud-based data lake on Azure.

I developed automated scripts to run on raw data to validate and cleanse the data for analysis. I also used Azure's cloud services such as Data Factory, Databricks, and Storage.

Once the data pipeline was built, I developed several dashboards on Looker (Google Data Studio) and Power BI based on the data to aid the business in decision-making.

I worked on other projects during my time here, developing web scraping scripts from flight tracker sites using Python to deliver daily reports to the business to help create macro research-based trading strategies for financial clients.

During the pandemic, I supported over 200 users in data science and data analytics functions on the AWS SageMaker machine learning platform within the UK government healthcare department.

Resolved complex problems relating to predictive analytics and visualizations for the Joint BioSecurity Centre in the UK. I helped build national alert dashboards based on trends of positive COVID-19 rates and deaths across the UK.

Developed an interactive Tableau dashboard for internal company users to aid with daily operations and insights.

A B2B fintech business that lends to companies on a short and medium-term basis. Key metrics included company loan utilization and loan performance.

Collaborated in a cross-functional team with engineers and a product manager to define content performance metrics and a supporting data model in Amazon Redshift and produce analysis and self-service dashboards on top of this data, including experiment analysis and tracking dashboards.

Tasked with refining and validating the supporting data model in Amazon Redshift. Used SQL to explore data and identify necessary changes, collaborating with engineers to implement changes.

Defined performance metrics based on exploratory data analysis, product considerations, and statistical power analysis, such as tweaking metric definitions to allow A/B tests to run faster. Built self-service experiment analysis and metric tracking dashboards using SQL and Python in a Sisense visualization product.

While at Expedia, I focused on data analytics and data science in search engine marketing and SEO for Hotels.com.

Challenged digital marketing strategies and improved growth by running incrementality tests such as pre and post-testing, A/B testing, and time series modeling. Found opportunities for collaboration between the Expedia brands to leverage the combined market share.

Used various predictive modeling techniques such as linear regression and other classification models to predict yearly revenues for newly onboarded hotels. Spent a lot of time in data cleansing and feature engineering using both Python and SQL.d

Developed several analytical dashboards using SQL and data visualization tools such as Tableau and Qlik Sense.

EmptyChair was an online marketplace via a website and iOS app focused on selling creative workshops and experiences around London, UK.

Co-founded the company as CTO and built the website and back end for the booking platform to run entirely automatically. The website was built on a LAMP tech stack connecting to Google Analytics for customer analysis.

The platform grew to over 10,000 users in the first year and registered over 200 partners.

Conducted a three-week analysis of churn rates of a monthly subscription-based product. Brought together multiple data sources spread across 3rd-party data integration platforms and internal data warehouses, including ChartMogul, Amplitude, and Amazon Redshift.

Explored key drivers of user cancellations, including user demographics, properties, and behaviors. Built a pipeline to assess feature importance using an XGBoost model and produced graphical charts to analyze patterns and correlations across cancellation rates and features.

Finally, I leveraged ChartMogul's churn rate calculations regarding the customer, net MRR, and gross MRR to build a coherent story around the development of the churn rate over time.

To improve hospital capacity management, the team and I developed a predictive model to identify how long a new patient at a hospital would likely need to spend in a hospital bed. This would be based on the patients' initial screening metrics and demographics, such as blood pressure and cholesterol levels.

This project was part of a "Health Hackathon" where doctors came together with data scientists and jointly developed a solution.

Utilized a linear regression model and classification models to give accurate predictions on time in hospital beds. Worked alongside medical doctors to incorporate useful features and ensure non-collinearity in the models.