Cloud & systems
FocusSelf-hosted delivery, SSR architecture, CI/CD, and reliability budgets for small-scale production sites.
- Next.js SSR / SSG delivery
- Docker + nginx reverse proxy
- CI deploy and health checks
- Performance and cost-aware hosting
Loading command center
Théo Marini · Systems engineer and technical builder
theom.fr
Investigates technologies, builds prototypes and production-oriented systems, tests constraints, and documents trade-offs.
Positioning
Théo Marini explores emerging technologies, builds working systems, tests their limits, and documents what creates real technical and operational value. Focus areas are backed by shipped work on this site; investigation domains are labeled as such.
Self-hosted delivery, SSR architecture, CI/CD, and reliability budgets for small-scale production sites.
Client-side WebGL scenes, scroll-driven interaction, and WebGPU/WASM labs kept off the critical SEO path.
Hardware/software prototypes, sensors, and edge constraints — explored as engineering interest, not claimed production portfolio yet.
Fintech infrastructure patterns and financial automation — tracked as investigation, not as shipped client work on this site.
Selected work
Production systems, stable demos, and prototypes — labeled by maturity. No fabricated client case studies.
production · active
Sole designer/engineer — architecture, implementation, hosting, and content.
Problem — Present engineering work with a distinctive immersive surface without making the site invisible to crawlers and answer engines.
Outcome — Live site on Hetzner with SSR pages, JSON-LD, llms.txt, bidaily SEO audit pipeline, and progressive WebGL enhancement.
experiment · stable demo
Implementation of R3F canvas, adaptive DPR, reduced-motion fallback, and scroll coupling.
Problem — Create a memorable visual field that remains optional for accessibility and does not block text LCP.
Outcome — Stable dither shader experiment in the repo; homepage hero evolved to a procedural command-center storyboard (walkway / platform / curved HUD screen).
experiment · stable demo
Architecture, procedural assets, quality profiles (mobile/desktop), bloom budget, reduced-motion poster.
Problem — Deliver the cinematic storyboard (wide → approach → HUD → global map) without blocking SSR copy or requiring a Blender GLB on day one.
Outcome — Live on homepage with mobile/desktop quality tiers, storyboard camera sampling, and WebGL/poster fallbacks.
experiment · stable demo
Crate + wasm-pack artifacts + lab page WebGPU demo with honest maturity labeling.
Problem — Separate systems-engine experiments from the design-critical Three.js hero so iteration does not break SEO delivery.
Outcome — Live demo on /labs/wgpu-engine: WASM engine identity plus visitor-GPU WebGPU draw; reduced-motion and no-WebGPU fallbacks.
Labs
Prototypes, benchmarks, rendering experiments, and Rust/WASM work that may not yet be production-ready — maturity labeled honestly.
Stable demo
Testing — Whether a scroll- and pointer-driven dithered WebGL field can brand a portfolio without blocking LCP or accessibility.
Why — Immersive differentiation must remain optional; crawlers and reduced-motion users still need the message.
Stable demo
Testing — A Rust→WASM control plane plus a minimal WebGPU draw path, isolated from the Three.js marketing hero.
Why — Compute-oriented labs need a systems stack; hero design velocity should not depend on that stack.
Work in progress
Testing — Compute-shader particle density and interaction cost on recent GPUs and phones.
Why — Validates whether WebGPU compute is practical for portfolio-scale demos without remote streaming.
Writing
Architecture decisions, SEO/GEO trade-offs, and engine comparisons — written to be citable.
webgl
Keep the wow on the client GPU, put brand and answers in SSR HTML, and lazy-load the canvas after text LCP so crawlers and humans both win.
Read note →geo
GEO structures your pages so ChatGPT, Perplexity, and AI Overviews can quote accurate answer capsules about you, your stack, and your labs.
Read note →rust
Use Three.js or R3F for marketing heroes and scroll storytelling; reserve Rust wgpu WASM for compute-heavy labs where a shared native/web engine matters.
Read note →Method
01
Map the problem, constraints, technologies, and failure modes before committing to an architecture.
02
Define interfaces, trade-offs, validation criteria, and what “good enough” means for the next iteration.
03
Ship a working implementation — prototype or production-oriented — that can be exercised end-to-end.
04
Measure performance, resilience, feasibility, and limits. Prefer observable outcomes over claims.
05
Record decisions, failures, and next steps so the work can be reused, cited, or challenged.
About
Systems engineer and technical builder based in Marseille, France. Builds and documents working systems at the intersection of self-hosted infrastructure, immersive web rendering, and Rust/WebAssembly labs.
Immersive web is a differentiator inside a broader systems practice: delivery, measurement, documentation, and honest experiment maturity.
Conversations about architecture, prototypes, and constrained engineering problems are welcome.