Thomas Lund

Simulated Legal SaaS Platform

Built a self-hosted legal-tech platform centered on structured document flows, product discovery, and simulated checkout journeys.
Designed a multi-page experience for browsing legal services, reviewing offerings, and navigating a realistic SaaS-style customer flow.
Created an internal agent-view interface to present the platform from both the user and operational perspective.
Integrated lightweight backend serving and on-page analytics components to support interactive product storytelling.
Deployed and maintained the project using Python, SQL, Docker, and frontend web technologies.

Python SQL Docker Streamlit

Streamlit Business Intelligence for Legal Commerce

Built an interactive Streamlit dashboard analyzing order trends, product performance, and customer lifetime value from a simulated legal e-commerce dataset.
Visualized weekly and monthly order development, top products by volume and revenue, and purchase timing patterns across days and months.
Identified cross-sell opportunities by surfacing the most frequently co-purchased product pairs in the same order.
Compared repeat customers vs. one-time buyers to quantify revenue contribution and order frequency behavior.
Deployed as a containerized Streamlit service integrated into the portfolio site with CSV-driven data pipelines.

Python Streamlit Pandas Matplotlib Docker

View GitHub

Collaborative modeling and systematic performance evaluation of complex datasets.

This project develops a probabilistic model to capture causal relationships in protein-signaling networks using Bayesian networks. It includes simulation of data, inference of hidden protein states, and parameter learning via logistic regression and EM algorithms. The approach enables understanding of how observed proteins influence unobserved ones and models biological dependencies efficiently.

Modeling Data Analysis Git

View GitHub

A high-performance deep learning solution for the German Traffic Sign Recognition Benchmark with an accuracy of 99.63%.

Designed and implemented a high-performance traffic sign classification model for the German Traffic Sign Recognition Benchmark (GTSRB), achieving 99.63% accuracy without transfer learning or ensembling. Built a custom Mini-ResNet architecture in PyTorch and trained it from scratch using advanced learning rate scheduling (warmup + cosine annealing). Developed a fully automated and reproducible training pipeline, optimized for efficient execution in Google Colab environments.

Deep Learning Python Computer Vision

View GitHub

End-to-end classifier for web domains, implementing data scraping, analysis, and predictive modeling[cite: 19, 25].

Built an automated real estate monitoring system using Python and Playwright to track Danish housing listings in real time. Implemented dual detection strategies (hash-based and URL-based) to accurately identify new properties across dynamic websites. Integrated Telegram alerts for instant notifications, ensuring fast response to new market opportunities.

API Python Automation

PROJECT_REGISTRY