Uses

Languages

• Python for AI, data and backends (FastAPI), TypeScript for everything that has a screen.
• Rust when the program has to be fast and honest: the game engine, the update server.
• Go for small, static, single-binary tools: collectors, sidecars, services with no runtime to ship.

Building

• FastAPI and Fastify on the server; React with Vite, TanStack Query and shadcn/ui on the client; Tailwind CSS when the design system is not this one.
• PostgreSQL by default, with PostGIS and TimescaleDB when geography or time series come into play; Redis for queues and caches; Qdrant for vectors.
• Bevy for the game, Tauri for desktop, DeepStream and TensorRT for vision on the edge.

AI

• LiteLLM in front of every model, so the application never learns a provider's SDK; Langfuse for traces; LM Studio to run the same stack fully local.
• Claude Code as a daily pair, with skills for the repetitive parts, and a sandbox of my own when an agent has to run unattended.

Operating

• A K3s cluster deployed by Argo CD from a git repository, Helm charts of my own, Jenkins building in pods, Sealed Secrets, WireGuard between the nodes, MinIO for objects (see the infrastructure page).
• Docker everywhere, Cloudflare in front (DNS, Pages for this site), Spark profiles before touching a game server.

Everyday

• Brave as the browser, Notion for notes and plans, Figma when something needs drawing before it needs code.