I am a Technical Lead & AI Solutions Architect. I turn real-world professions and workflows into software systems by starting with their day-to-day reality, their real frictions, and their canonical references. The stack comes second.
My path from biology through regulated-system validation to agentic delivery shaped a simple operating method: analyse, structure, orchestrate — then prove. Automation handles the time-consuming work; architecture, arbitration, and validation stay human.
The work → domain understanding · software architecture · agentic delivery
The method → bounded missions · governed sources · blocking gates · live evidence
Main languages → Python · Go · TypeScript · C · PowerShell · Shell
| Principle | In practice |
|---|---|
| Understand the domain | Start from the profession, its constraints, and what actually happens day to day — not from a fashionable stack. |
| Structure the promise | Turn sources and decisions into explicit contracts, traceable data, deterministic rules, and versioned artifacts. |
| Orchestrate bounded work | Give agents a defined objective, tools, limits, definition of done, and evidence to return. The model is interchangeable; the method is not. |
| Prove the result | Use blocking CI, real acceptance flows, and honest project states. Static presence and mocked tests do not prove a live outcome. |
Five public repositories explore how source-driven and agent-assisted work can remain traceable, bounded, and verifiable:
| Repository | Role | Public state |
|---|---|---|
| NOMOS | Turns governed source material into traceable canonical nodes and auditable artifacts | v0.1.0-ALPHA; bounded proof of concept |
| RECURSIO | Collects authorized, versioned corpora and prepares NOMOS-compatible exports | Initial Go CLI and database migrations; complete pipeline still planned |
| ORDO | Coordinates bounded multi-agent delivery and gates merges on evidence | Toolkit published; release-validation gates remain open |
| PRAXIS | Reconciles code and features, executes acceptance scenarios, and reports evidence | v0.2.0; explicitly pre-1.0 |
| TECHNAI | Capitalizes skills, references, and safety rules across agent runtimes | Core library and synchronization work; optimization layer remains roadmap |
|
Swiss architecture workflows. An address becomes sourced parcel constraints under a sourced or unknown contract. Actions follow inspect → dry-run → scoped approval → execute → registry, so nothing is applied without explicit human approval.
|
A living tabletop ruleset. Canonical sources feed structured catalogs; a deterministic rules engine is the next phase. The design keeps future AI narration separate from dice, damage, and threshold calculations.
|
|
Classical-music mastering. A minimal, capped processing chain preserves dynamics, targets EBU R128, and uses null tests and A/B comparison to make transparency measurable.
|
Algorithmic trading. Strategies are described, backtested against history, and executed in simulation by default; live connections require deliberately supplied credentials.
|
- Scintilla — pure-C embedded workshop with hardware-independent logic, host-side tests, sanitizers, and AVR/ATmega328P and ARM7TDMI-S adapters.
- MemoForge — bilingual, zero-dependency puzzle game for learning C pointers and memory by assembling programs that run.
- Umbra — offline prototype of an isometric hack-and-slash roguelite; the vertical slice is still in progress, and the older standalone
umbra-*repositories are superseded and archived. - Meetinity — professional-networking platform with existing services and front ends; the matching core remains unfinished and the project is paused.
- Windows 11 Theming Suite — native theming through Windows APIs, PowerShell automation, a desktop GUI, and C++ components, including one opt-in path that uses an undocumented composition API.
- Universal Project Launcher — per-project context profiles and one-click assisted-development launches across Windows, Linux, and WSL.
Understanding software as a profession means returning to its foundations:
- 42 Training — Linux-first C, shell, and logic practice, extended into a constrained three-track learning platform.
- Piscine Prep — versioned curriculum, Bash training engine, local tests, review workflow, and progress tracking.
- revision42 — a deliberately small C revision workspace.
