Creator of EIDOS · Local AI · Multi-Agent Systems · MCP · Workflow Automation · Computer Use
Ajdovščina, Slovenia · Open to remote / EU opportunities
Portfolio · Repository map · Work with me · All repositories · Stars
I am an AI-native builder focused on turning ideas into working systems through AI-assisted engineering.
My work combines local language models, coding agents, automation, APIs, browser/desktop tooling, persistent memory, knowledge graphs and verification loops. I use AI as an implementation collaborator while I define the product behavior, architecture, constraints, workflows and acceptance criteria.
My main project is EIDOS, a local-first persistent agent runtime exploring long-lived memory, reusable capabilities, multi-agent coordination, controlled computer use, action-effect verification and rollback-oriented self-repair.
I am especially interested in systems where AI does more than answer questions: it can use tools, observe effects, retain state, coordinate work and improve repeatable workflows.
EIDOS is my main research and product project.
It explores a system architecture built around:
- persistent memory and graph state;
- multi-agent / Colony deliberation;
- one arbitration layer for final decisions;
- browser and desktop interaction;
- action → observed effect → verification loops;
- reusable capability learning;
- local model and tool adapters;
- isolated self-edit and rollback research;
- a dedicated Desktop 2 workspace for agent interaction without taking over the user's primary desktop.
A convincing answer is not the same thing as a verified effect.
The public repository deliberately separates implemented, wired, executed, measured and reused capabilities instead of treating documentation as proof.
A guided automation toolkit for the OnePlus 7 Pro GM1913 covering recovery-oriented workflows, Android/Windows ARM preparation, Kali NetHunter tooling, backup/verification and device setup.
This project reflects another part of how I work: combining existing low-level tools into a safer, reproducible workflow for a real device.
The public source release includes Python and Bash launchers, a Spanish setup guide and firmware-presence checks. Users provide their own compatible firmware and device-specific backups. Read the source guide →
I built this automation for Martin Miklavčič in Slovenia, founder of Miklavčič marketing, for use with one of his clients.
A Python application that turns DOCX articles and photographs into structured Markdown, SEO metadata, FAQs and quick facts. It supports review in a local browser, standalone HTML export and optional draft/publication operations through a compatible CMS API configured by each user. The public edition includes synthetic examples and keeps client accounts and documents private.
Source code, setup and usage →
A client storefront for Petra + Erik / Casa de Té, using the latest improved version selected for publication. It combines seven 3D scenes with a classic shop, Spanish/English/Slovenian interfaces, catalog filters, botanical product notes and a shared shopping cart. Checkout produces a local demo receipt; purchase and contact actions are simulations. The 3D view requires WebGL; the classic storefront is available directly below.
Open the demo → · Open the classic shop → · Source and project notes →
A reviewed archive of the 2026-09-09 snapshot, kept separate from current EIDOS. It contains a durable SQLite task queue with dependencies and recovery, Markdown/FTS5 memory with provenance, and an HTTP/SSE client for local models.
A synthetic offline demo and component tests document what can be exercised independently. Autonomous orchestration remains disabled. The repository also documents the snapshot's unresolved original licensing and attribution.
Explore the archive and offline demo →
Two newly prepared Python utilities: a local metadata inventory that reports aggregate counts without names or paths, and an offline analyzer for synthetic authentication-response summaries. The release includes reproducible fixtures, an anonymized static case study and attributed upstream references. It also preserves one recovered generic wireless workflow as an experimental archive, disabled by default, with its original provenance and unresolved licensing documented.
This publication combines new defensive tools and documentation with one selected historical source file. The wider local research collection was not imported. Explore the lab →
Current areas of focus
- AI automation and internal tools
- agentic workflows and multi-agent systems
- local/private AI
- persistent memory and knowledge graphs
- browser and desktop computer use
- MCP / tool integration
- rapid AI product prototyping
- Linux automation
- Android / Windows-on-ARM experimentation
- verification, rollback and safer autonomy
I keep a broad GitHub workspace because I actively study and compare tools across AI, automation, operating systems, security and agent infrastructure.
| Repository | Role |
|---|---|
| EIDOS | Flagship persistent local-agent runtime |
| msmKALI-automatization- | OnePlus 7 Pro recovery / Windows / NetHunter workflow automation |
| Martin client automation | Workflow automation developed for Martin Miklavčič for a client of his; public reusable edition |
| Petra tea shop | Client storefront project with an interactive public demo |
| ARCA / COLMENA | Reviewed historical research archive; synthetic demo and independently tested components |
| Security Research Lab | New offline utilities, defensive analysis and a guarded experimental wireless archive |
| EIDOS-OFFICIAL | Historical EIDOS repository; preserved as a redirect/archive |
These repositories are kept as forks, references or research material. They are not presented as my original authorship.
UI-TARS Desktop · Neural Memory · ClawTeam · OpenClaw · SkillOpt · MiMo Code · CodeGraph · ToolNeuron / Android AI · PrivateLM · Ollamacode CLI
Windows for OnePlus · WOA Tools · QuickRotate WOA · Rucky / NetHunter
My account also contains security and OSINT tooling used for learning, comparison and authorized research. A repository appearing in this account should not be interpreted as original authorship unless explicitly stated.
My workflow is deliberately verification-oriented:
idea
↓
specify behavior
↓
use AI / agents to implement
↓
run it
↓
observe the real effect
↓
verify
↓
document
↓
keep, improve or roll back
I care about the difference between:
- code existing;
- code actually being connected;
- a path being executed;
- a measurable effect being observed;
- a learned capability being reused later.
That distinction is central to EIDOS and to how I evaluate AI-generated software.
I am interested in full-time, contract and freelance work involving:
- AI Automation Specialist
- Agentic Systems / AI Agents
- Generative AI implementation
- AI workflow automation
- internal AI tools
- local/private AI
- rapid AI product prototyping
- AI-assisted operations and process automation
Remote · Slovenia · EU
If your team has a manual process that could become an AI-assisted workflow, an agent that needs tools and memory, or an AI prototype that needs to become a working system, that is the kind of problem I want to work on.

