I build AI products and the systems behind them. Over thirty years in software, hardware, and R&D have taken me from low-level systems to enterprise architecture, technical leadership, and startup mentorship.
I run Gedank Rayze in Lisbon, where my work includes AI product architecture and data forensics for enterprise clients in regulated industries. My background in digital forensics, including work at Cellebrite, shapes how I build: data should be traceable, decisions auditable, and systems reliable under pressure.
Our AI work rests on three pillars: IronCrew for agent coordination, IronFlow for workflow execution, and CogniGraph for connected data.
IronCrew — Agents Under Human Command
A Rust runtime for teams of AI agents, with agents, tasks, and tools defined in Lua. It coordinates parallel work, tool calls, and shared memory, with explicit checkpoints for human input and approval. Deploy it as a native binary with your choice of model provider.
IronFlow — Deterministic Workflows
A workflow engine built in Rust with Lua scripting for ETL, API integrations, CI/CD, and task automation. Dependency graphs, parallel execution, retries, and a REST API make execution explicit. Ships as a single binary without a Python or Node.js runtime; PDF rendering nodes additionally require Pdfium.
CogniGraph — Connected Data and Evidence
A database built in Rust that combines JSON documents, typed graph relationships, full-text search, and vector search in one process. Query connected records with CGQL and keep relationships linked to their source material — a foundation for AI retrieval and forensic analysis.
AI product architecture · Enterprise systems · R&D management · Technical leadership · Data forensics · Cloud & DevOps · IoT
| Project | Description |
|---|---|
| SPLADE Model Trainer | Train, evaluate, and deploy models for sparse information retrieval |
| Memsplora / SPLADE REST API | In-memory SPLADE content server with FAISS search |
| tiny-crew | Multi-agent AI system for collaborative task execution |
| openai-decorators-and-tools | Patterns for building OpenAI-compatible function-calling tools |
| llm-chatbot | Chat UI examples across multiple LLM providers |
| gemini-ocr | PDF screenshot analysis using Google Gemini |
| Project | Description |
|---|---|
| Bun File-System Router | TypeScript server template using Bun's native file-system routing |
| FastAPI | Dynamic route loading for FastAPI |
| Sanic | Dynamic route loading for Sanic |
| Flask | Dynamic route loading for Flask |
| Tarantool Starter | Minimal HTTP service for Tarantool |
| Tarantool Jobs Queue | Showcase of job queue management through a REST API, built with Tarantool and Lua |
| Project | Description |
|---|---|
| microflow | Python workflow engine with task dependencies, retries, and JSON or Redis state persistence |
| Project | Description |
|---|---|
| NLUI Demo | Natural-language interface using React, Ant Design X, and OpenAI, with server-controlled charts, tables, forms, and actions |
| Dutchy Design System | UI kit inspired by Dutch graphic design, built with Bun SSR and server-only JSX components |
| Project | Description |
|---|---|
| lua-arangodb | ArangoDB client written in Lua for OpenResty |
| arangodb-haystack | Haystack components for working with ArangoDB |
Open to conversations about AI product architecture, technical leadership, and collaboration. Connect with me on LinkedIn or explore my work at skitsanos.com.





