about

AI and agentic systems, engineered end to end

Agentic AI & Knowledge Systems is the work: agent harnesses and orchestration, evaluation that makes quality reviewable, retrieval and agent memory, and models running on hardware you control — engineered end to end.

get in touch read the writing

27
articles: 70
tutorials reviewed: 05
courses: 10
tools in the registry

focus

What I build

01// the frontier

AI & agentic systems

Machine learning and agentic AI engineered end to end — the work now pointed at systems that think.

agent harnesses, orchestration and tool use
context engineering — what the model sees, and when
evaluation: eval suites and LLM-as-judge, so quality is a statement someone can review
shipped open tools, not demos

the published tools →

02// the stack underneath

AI systems

The part that has to hold: data in, retrieval working, quality measurable, running on infrastructure you can name.

machine-learning systems and pipelines
knowledge graphs, RAG and semantic retrieval
agent memory that survives the session
local and self-hosted inference — open-weight models on hardware you control

the learning archive →

agentic AI & agent orchestration

machine-learning systems & pipelines

retrieval, knowledge graphs & agent memory

background

How the AI work began

TU Wien architecture → parametric design → BIM → BIM management → data engineering → AI engineering → agentic AI.

Twenty years of structuring large, semi-structured data under standards-driven governance — then learning to build the systems that query it.

Machine learning and AI engineering, built on the design–data bridge; today: agentic systems — harnesses, orchestration and evaluation — in the public sector.

depth

Where the weight sits

agent harnesses & orchestration
evaluation & quality — eval suites, LLM-as-judge
retrieval, GraphRAG & agent memory
local & self-hosted inference
data engineering & pipelines
deployment & operations

toolkit

The stack, by domain

agentic ai / ml// the frontier

Hermes Agent
agentic AI
AI engineering
machine learning
marimo

languages// the craft

Python
TypeScript
JavaScript

frontend / design// the interface

React
web design
graphic design
Flutter

infra / devops// the backbone

Docker
Linux
Traefik
self-hosted

data// the insight

data engineering
data science
retrieval
embeddings

boundaries

What I do — and what I don't

Bounded artifacts, handed over as working systems — not retainers, not headcount, and no open-ended outsourcing.
Not a software shop: no full custom build with a maintenance contract, no takeover of an existing codebase, no permanent development capacity.
Not a body shop: no staff augmentation, no leasing of my time by the day, no day rate without a ceiling.
Not a legal practice: regulatory statements are dated and sourced, and anything that has to be signed as legal advice comes from your lawyer.
Independence is structural — no commissions, no licence resale, no vendor whose product I have to move.
Handover by default: the system runs when the builder is gone.

manifesto

Software should read like a good shell session: every line intentional, every output honest, nothing hidden behind a dashboard.

location

Vienna, Austria

Everything here is written and maintained by one person, in Vienna. The archive currently holds 27 articles, 70 tutorials, 5 courses and 10 tools — shipped from a terminal, not a marketing site.

get in touch read the writing

48.2082° N · 16.3738° E

based in: Vienna, AT
working: public sector
focus: agentic AI · AI systems
elsewhere: github.com/mokmail