APIs, AI runtimes, Linux servers, and data pipelines.
Architecture · Projects · Stack · Activity · LinkedIn
I write Python, TypeScript, Rust, and C#. My work includes APIs, background workers, Linux services, native desktop runtimes, live data paths, and operational tooling.
I started in data science and moved toward backend and platform work. I still use that background for model evaluation, data quality, and reproducible pipelines.
I spend most of my time on service boundaries, VPS and dedicated servers, event delivery, model serving, health checks, and recovery.
The core runs on a VPS or dedicated host. It serves snapshots and ordered events from the same runtime; state and lifecycle controls stay outside the delivery path.
Design notes
- Ordered events carry live changes. Snapshots handle startup, recovery, and degraded operation.
- External adapters normalize source-specific payloads before domain code sees them.
- Every queue, cache, history, worker pool, and retention window has a fixed limit.
- Health, maintenance, read-only mode, and shutdown stay available when live workers are paused.
- Live and test runtimes keep separate state, caches, and secret namespaces.
- New artifacts are verified before they replace the active version.
| Project | System |
|---|---|
| SCG Infrastructure | Supervised Rust control-plane node with revisioned SQLite state, ordered SSE events, deterministic service lifecycles, and isolated runtime environments. |
| NLP Sentinel | FastAPI inference API: typed schemas, cached model loading, file-based or Azure telemetry, and a feedback endpoint. |
| Spark Vision | PySpark image-feature pipeline using pandas UDFs and MobileNetV2, with Parquet output on EMR/S3. |
| Reco Engine | Azure Functions recommendation API with Blob-backed artifacts, an in-process cache, and a cold-start fallback. |
| Urban Segmenter | Keras segmentation pipeline served by FastAPI, with a Streamlit client that runs in-process or against the API. |
More engineering work
- Auto-CV — Python desktop application built around SQLite repositories, deterministic document generation, and a managed
llama.cppprocess. - Neural Exchange — PyTorch time-series experiment with causal baselines, saved model artifacts, generated reports, and offline tests.
- Freight Network — Python graph-analysis package using deterministic synthetic data, NetworkX metrics, generated reports, tests, and CI.
| Layer | Stack |
|---|---|
| Languages | Python · TypeScript · Rust · C# · SQL · JavaScript · PowerShell |
| Backend and APIs | FastAPI · Uvicorn · Fastify · Express · Pydantic · REST · SSE · typed IPC |
| Systems and desktop | Cargo workspaces · Tauri 2 · .NET / WinForms · Node.js · native launchers · process lifecycle |
| AI and data | PyTorch · TensorFlow · scikit-learn · llama.cpp · pandas · PySpark · Parquet · NetworkX |
| Web and rendering | React 19 · Svelte 5 · Vite · PixiJS · MapLibre GL · Canvas · Streamlit |
| Storage and integrity | SQLite · rusqlite · Serde · JSON · SHA-2 · filesystem adapters |
| Infrastructure | Linux · Docker · VPS · dedicated servers · OVHcloud · Hetzner · AWS EMR/S3 · Azure Functions/Application Insights |
| Quality and delivery | pytest · Ruff · Cargo test · Clippy · Vitest · Playwright · ESLint · GitHub Actions |
LinkedIn is the best way to reach me.