Turn Data Into Decisions That Drive Revenue
I design and build decision-ready Power BI dashboards that
teams can use immediately.
From messy data to clear KPIs, trusted numbers, and
confident decisions.
Current Projects
A selection of my recent open-source work and public project contributions, hosted on GitHub.
Vantage Alpin: Analytics Engineering
Production financial reporting engine replacing legacy PDF reports with a dynamic, decision-ready Power BI system.
Concept
This project acts as the central financial reporting engine for Vantage Alpin. It replaces legacy PDF reports with a dynamic Modern Data Stack (MDS) that processes stochastic transaction data into a clean Star Schema.
Features
- Controlled Stochasticity: Uses Non-Homogeneous Poisson Processes (NHPP) to model realistic demand patterns.
- Product Economics: Pricing follows Log-Normal distributions; popularity follows Pareto (Power Law).
- Marketing Attribution: Algorithmic allocation of ad spend to order lines based on revenue share.
Architecture
- Star Schema: Central
fct_transactionstable linking to Type 1 SCD product dimensions. - Currency Normalization: Robust handling of multi-currency transactions (EUR/CHF) with daily rate lookups.
- DirectQuery-like Extract: Custom Parquet/CSV export workflow from DuckDB to Power BI for performant local development.
End-to-End SaaS Operating System
A full-stack simulation engine that translates static business plans into living data systems.
Concept
This project implements a "Rapid Data Function"—instantiating an entire end-to-end data stack (Stochastic Simulation, Data Warehouse, Algorithmic Forecasting, and BI) purely from a static business definition.
Features
- Stochastic Engine: Simulates Customer Acquisition (Poisson), Retention (Hazard/Survival), and Revenue expansion.
- Regime-Aware Forecasting: A Prophet model that detects "Growth Regimes" to avoid overfitting on early-stage data.
Architecture
- Contract-First Engineering: The entire simulation scales dynamically based on a JSON contract extracted from the Excel Operating Plan.
- Medallion DWH: A rigorous dbt project structure (Bronze/Silver/Gold) ensuring clean lineage.
- Agentic Build Pipeline: Treating business logic as code, we use cursor/windsurf to scaffold the entire stack from specs.
Interactive Yield Curve Heatmap
An interactive web application encoding 25+ years of German sovereign yield curve data into a single, high-performance canvas heatmap.
Concept
A web application that visualizes 25+ years of German sovereign yield curve data in a single view. The application fetches monthly Svensson zero-coupon yield estimates directly from the Bundesbank SDMX REST API.
Features
- High-Performance Rendering: Uses HTML5 Canvas and ImageData for pixel-level bilinear interpolation, providing smooth and accurate visuals.
- Interactive Controls: Configurable date and maturity ranges, 26 color palettes, and interactive tooltips for precision exploration.
- Data Integration: Live data fetching with automatic fallback to a synthetic Nelson-Siegel model dataset if the API is offline.
Architecture
- Frontend: React 18 for UI and D3 (v7) for color interpolators and scales.
- Export: On-device high-resolution PNG rendering via an offscreen 2560x1440 canvas.
B2B Pipeline Forecasting Prototype
Business-day sales pipeline forecasting prototype with competing outcomes, age-based conversion kernel estimation, and PIT reconstruction.
Concept
Model how a B2B opportunity pipeline converts over time (in business days), then forecast future daily conversions by combining expected conversions from existing active stock and newly arriving opportunities. Built as a transparent, auditable forecasting framework.
Features
- Kernel Estimation: Competing-risk kernel estimation (won, lost, censored) from a trailing training window.
- PIT Reconstruction: Robust Point-in-Time reconstruction from SCD2 status history to prevent look-ahead leakage.
- Synthetic Data: Generator includes realistic behaviors like business-day seasonality, weekend suppression, and regime switches.
Architecture
- DuckDB Core: A fast analytics warehouse managing calendar, opportunity facts, and SCD2 status history.
- Evaluation Framework: Daily error metrics (MAE/RMSE/WAPE) and rolling multi-as-of backtesting with age-attributable error surfaces.
- Multi-panel Visualization: Structured forecast architecture rendering existing-pipeline and future-arrivals convolution heatmaps.
ShaderGradient Playground
A browser-based, real-time shader gradient simulator built on Three.js and custom GLSL for rendering and exporting animating mesh gradients.
Concept
The playground renders a live, animating gradient mesh using a custom simplex-noise vertex shader. Dial in animated mesh gradient visuals—spheres, planes, water planes—and export them as looping GIFs or videos directly from the browser without any installation.
Features
- Real-Time Shading: 3 customizable gradient colors blended across geometry based on noise, full lighting/environment controls.
- Geometry & Camera: Tweak geometries (sphere, plane, etc.), noise displacement, and camera angles freely.
- Rich Export: Export looping animations as GIFs or videos (WebM/MP4 up to 4K), or copy the exact state as a React JSX snippet.
Architecture
Built purely in the browser using Three.js and custom GLSL. It features an integrated preset system with localStorage, and client-side media encoders (omggif and WebCodecs/MediaRecorder API) for direct asset generation.