Software Architect | Data Scientist | Ph.D. Mathematics
I design high-performance data systems in Rust and C++ and apply statistical and machine-learning methods to industrial planning problems. My work spans DuckDB extensions, time-series forecasting engines, and GenAI infrastructure (RAG, MCP servers, foundation model inference).
LinkedIn • Crates.io • DataZooDE
- Time-Series Forecasting -- Hierarchical, probabilistic, and intermittent-demand forecasting for supply chains, built as native DuckDB extensions.
- Statistical Computing -- Production-grade regression, hypothesis testing, and causal inference in Rust, exposed through DuckDB and Polars.
- GenAI & RAG Infrastructure -- Vector databases (HNSW/DiskANN), retrieval-augmented generation, and Model Context Protocol (MCP) servers for AI-assisted development.
- Foundation Model Inference -- Pure-Rust inference engines for time-series models, targeting edge and WASM deployment without Python dependencies.
- Enterprise Data Integration -- DuckDB extensions for SAP and API ecosystems, bridging legacy ERP systems with modern analytical workflows.
- Inventory & Supply Chain Optimisation -- Stochastic inventory models and demand planning applications.
- Forecast Accuracy & Speed — anofox-forecast delivers 2,900x faster forecasting, enabling near-real-time demand planning that reduces stockouts and excess inventory.
- Inventory & Working Capital — Stochastic inventory models optimise safety stock levels, freeing working capital while maintaining service levels.
- Enterprise Data Accessibility — erpl and flapi turn locked-away SAP/ERP data into queryable, API-accessible datasets, cutting integration timelines from months to days.
- Data Quality & Trust — Automated anomaly detection and validation (anofox-tabular) catches data issues before they reach dashboards and decisions.
- AI-Ready Infrastructure — RAG pipelines and vector search (Magpie) ground LLM responses in company knowledge, reducing hallucination and making GenAI safe for enterprise use.
- Reduced Infrastructure Cost — Pure-Rust inference (Chronos-2) and high-performance libraries (motif-rs, oxits-rs) eliminate Python overhead, cutting cloud compute costs and enabling edge deployment.
| Project | Highlight | Stack |
|---|---|---|
| flapi | DuckDB-powered API gateway with MCP server and VS Code extension | C++, DuckDB |
| erpl | DuckDB extension bridging SAP systems via RFC | C++, DuckDB |
| dbt-lineage-viewer | Fast CLI for visualising dbt model lineage | Rust |
| anofox-tabular | Anomaly detection, validation, and data preparation in DuckDB | C++, DuckDB |
| Project | Highlight | Stack |
|---|---|---|
| anofox-forecast | 2,900x faster than statsmodels; DuckDB community extension | C++, Rust, DuckDB |
| Chronos-2 | Pure-Rust re-implementation of Amazon's Chronos-2 time-series foundation model | Rust, Candle |
| oxits-rs | Time series classification and transformation library -- port of pyts | Rust |
| motif-rs | High-performance matrix profile library; 3--63x faster than stumpy | Rust |
| Project | Highlight | Stack |
|---|---|---|
| Magpie | Vector DB and RAG engine with HNSW, hybrid retrieval, AST-aware chunking | Rust |
| Project | Highlight | Stack |
|---|---|---|
| polars-statistics | High-performance statistical testing and regression for Polars | Rust, Python |
| Inventory Optimisation | Stochastic inventory models for demand planning | Rust |
| fdars | FDA algorithms -- depth measures, clustering, smoothing, regression | Rust, R |
flapi and erpl are DataZooDE projects.
| Package | Description |
|---|---|
| fdars-r | Functional Data Analysis R package with Rust backend |
| eventstudy | Financial event study analysis |
| case-based-reasoning | Case-based reasoning using machine learning methods |
Core:
Data & ML:
Infrastructure:
