PROJECTS.

I implemented the complete working application: the Vue frontend, search flow, Realty API integration, property views, map, authentication handling and final interface. The thesis and evaluation were completed as a four-person bachelor project.

Professional property data is accurate but dense and difficult to scan. We wanted to prove that the same data could be made easier to understand without sacrificing detail or access control.

Completed in May 2026. The delivered prototype includes Ambita SSO, property search, a structured property overview, ownership, encumbrances, buildings and location data. The thesis evaluation produced a SUS score of 87.5.

Demonstrate a clearer way for authorised professionals to understand complex property information and give Ambita a solid prototype for further product development.

I am designing and building the complete tool across state modelling, equity simulation, machine learning, decision rules, the desktop interface and the screen-vision pipeline.

The project lets me combine poker theory, probability, machine learning and computer vision in one measurable system — and compare model-based recommendations with transparent rules.

The working prototype has manual input, a Monte Carlo equity engine, a baseline PyTorch MLP, hybrid GTO recommendations, calibration and OCR, plus JSON hand logging. Current work focuses on validation, tuning and reliability.

Build a reliable offline research and hand-analysis assistant that can explain its recommendations and help study No-Limit Texas Hold'em decisions responsibly.

I will lead the project from workflow research and system design through implementation, evaluation and documentation.

I want to explore where AI can remove repetitive work while still keeping people in control of important decisions and final outputs.

The project is currently in discovery and scoping. Hands-on implementation is planned to begin in Autumn 2026.

Create a small set of reusable, auditable automations that connect models and tools to real workflows, with clear validation and human review points.

I built the complete personal analytics pipeline: game fetching, classic and duel normalisation, geographic enrichment, performance metrics, coaching reports, dashboard and per-round review tool.

Individual GeoGuessr mistakes are memorable, but recurring weaknesses are hard to see. I wanted an evidence-based way to review my history and identify what I should practise next.

The pipeline already produces normalised JSON and CSV data, aggregate metrics, a coaching report, a self-contained dashboard and a deterministic round coach. I am continuing to improve coverage and the depth of its coaching insights.

Turn raw game history into a self-contained personal coach that pinpoints geographic weaknesses, explains costly misses and makes practice more focused.