01  /  AI & Agentic Systems Engineer

MikelKrasniqi

I build production AI systems — autonomous agents, multi-model orchestration and applied LLM products — end to end.

View workRequest CV

Senior engineer with a rare builder's arc — goldsmith to CNC and robotics to full-stack to applied AI — shipping agentic systems across desktop, web and infrastructure, backed by 20+ years of leading teams and running businesses.

Now

Currently driving agentic-AI adoption inside a 100–150-person enterprise software company, while building production agent systems of my own.

02  /  What I do

Production AI, end to end

I design and build production AI systems — and help teams put them to work.

  • 01Autonomous, multi-agent runtimes with human-in-the-loop control
  • 02Cost-aware orchestration that routes across many models — cloud and local
  • 03Retrieval-augmented memory (RAG) and vector search
  • 04Local-first, privacy-preserving inference
  • 05Applied LLM products, taken from idea to shipped
  • 06Team enablement — mentoring engineers and driving agentic-AI adoption

03  /  Selected work

Systems I’ve built

A spread of my own work — from an autonomous multi-agent OS to shipped products and hands-on hardware.

01

Autonomous Multi-Agent OS

Production

A desktop operating system for AI agents — running autonomous, multi-agent workflows locally and privately on your own machine.

  • Hierarchical agents — a coordinator that auto-staffs specialised sub-agents — with human-in-the-loop approval gates that survive app restarts via SQLite checkpointing.
  • Cost-aware model routing that automatically picks the cheapest capable model across OpenAI, Anthropic, Gemini, Groq, Mistral and Ollama, with fully local inference as a fallback.
  • Local-first privacy: an AES-256-GCM encrypted vault, on-device vector memory and a spatial memory design — no cloud required to operate.
  • Production-grade engineering: hundreds of automated tests, CI, and shipped as a single signed desktop binary for Windows, macOS and Linux.
  • Tauri
  • Rust
  • React
  • TypeScript
  • LangGraph
  • LanceDB
  • llama.cpp
02

Multi-Channel AI Content Platform

Production

An agentic, multi-model web platform that writes a post once and adapts it for nine social channels.

  • A master orchestration agent routes work to specialised text and image sub-agents to produce platform-optimised posts end to end.
  • Provider-agnostic LLM layer with vector memory for retrieval-augmented context (RAG-ready).
  • Per-request token-cost tracking and graceful degradation when a provider fails.
  • Laravel
  • PHP
  • Agent framework
  • OpenAI
  • Gemini
  • Alpine.js
  • Tailwind
03

Deployed Consumer Product

Shipped product

A polished, shipped product for sending beautifully themed invitations and tracking the responses.

  • A genuinely shipped product: passwordless auth, content moderation, Stripe checkout via webhooks and an admin dashboard.
  • Built on a modern, fully typed stack (Next.js App Router, Prisma ORM, Postgres) with unit tests.
  • Next.js
  • React
  • Prisma
  • Stripe
  • PostgreSQL
04

Vertical SaaS Suite

Engineering

A family of vertical SaaS applications on a shared, reusable stack — across several different domains.

  • Breadth at scale: roughly a dozen domain-specific apps (time-tracking, event planning, design tooling and more) on shared infrastructure patterns.
  • Several integrate LLMs directly — agent-based features and model-powered generation.
  • A consistent Docker-based deployment and design-system approach across the whole family.
  • Laravel
  • Next.js
  • Docker
  • Google Gemini
05

Hardware & Robotics

Engineering

A decade of hands-on engineering before software — CNC machines, robotics and electric vehicles.

  • Designed and built a 3-axis CNC milling machine as a personal innovation (1000 × 1000 mm work area).
  • Contributed to the MV-4 mine-clearance robotics programme at DOK-ING, including chassis construction and team coordination.
  • Worked on the LOOX electric-vehicle project — mechanical construction and prototyping.
  • CNC
  • CAD
  • Mechatronics
  • Robotics

04  /  The path

Craft → code

An unusual route into AI engineering — and the reason I think about systems at every layer.

  1. 1995

    Goldsmith

    Trained goldsmith — precision craft and a maker’s eye for detail. The origin point of an engineer’s discipline.

  2. 2012

    Robotics & CNC

    Robotics and automotive work at DOK-ING and Dalekovod (Croatia) — including the MV-4 mine-clearance programme — and a self-built 3-axis CNC milling machine.

  3. 2013

    Operations & team lead

    Logistics & shop manager at Festamo (Germany): running online-shop operations and B2B, leading a small team.

  4. 2016

    Into software

    Fachinformatiker / Application Development (IHK) at the Macromedia Academy, Munich — and a development internship at BMW Group, building a Java/Spring payments tool for BMW Bank.

  5. 2018

    Applied AI

    Software & AI engineering at a 100–150-person enterprise software company — and four years building applied-AI products of my own, full time.

05  /  Capabilities

The toolkit

AI / LLM & Agentic

  • Agentic workflows
  • Multi-agent orchestration
  • RAG & vector memory
  • Tool-calling
  • Cost-aware model routing
  • Local inference (llama.cpp)
  • Fine-tuning
  • OpenAI · Anthropic · Gemini · Mistral · Llama

Languages & Frameworks

  • TypeScript
  • PHP
  • Python
  • Rust
  • Java
  • Laravel
  • React
  • Next.js
  • Astro
  • Vue
  • Angular

Infrastructure & Data

  • Docker
  • Kubernetes
  • FastAPI
  • Tauri
  • PostgreSQL
  • MySQL
  • MongoDB
  • Redis
  • RabbitMQ
  • LanceDB

Ways of working

  • Team leadership
  • Mentoring & enablement
  • Technical training
  • Architecture & design
  • Agile / Scrum
  • End-to-end delivery

06  /  Education & background

Trained, then self-driven

  1. 2016 – 2018

    Fachinformatiker — Application Development (IHK)

    Macromedia Academy, Munich

    Formal software-engineering training and IHK certification.

  2. 2013

    Metalworker / Locksmith (Schlosser)

    Vocational school, Zagreb

    Precision metalwork and fabrication.

  3. 1996 – 1998

    Goldsmith

    Vocational school, Zagreb

    Where the maker's eye for detail began.

Certifications

  • PPEDV — .NET Modern Architecture & Design Principles (2023)
  • Sololearn — Java, jQuery, HTML / PHP / SQL (2017)

Languages

German
Fluent — written & spoken
English
Fluent — written & spoken
Croatian
Native
Albanian
Native

07  /  Leadership

Not just a coder — a builder of teams

A genuine, multi-decade leadership record: 15+ years leading teams and running businesses as a managing owner across several companies, now applied to engineering — mentoring developers and driving technical adoption of agentic AI.

20+ yrs
leading teams & running businesses
Founder
managing owner of several companies
Teams
led groups of up to six
Enablement
mentoring engineers & driving AI adoption
Portrait of Mikel Krasniqi

08  /  About

From metal to machines to agents

My path is unusual. I started as a goldsmith, moved into CNC machining and robotics, retrained as a software engineer at the Macromedia Academy in Munich (Fachinformatiker, IHK), and for the last four years I've built applied-AI products full time. That range — from hardware to autonomous agents — is exactly how I think about systems: at every layer, end to end, with a builder's bias for shipping.

09  /  Contact

Let’s build something that ships.

mikelkrasniqi@gmail.comGitHub

Full CV (EN / DE) available on request — just send me an email.