Johannes Köppern
Portrait of Johannes Köppern

Dr.-Ing. Johannes Köppern

From control engineering research to AI products — Founder & CTO, narratiQ.

EmailLinkedInBook a call

Highlights

patents filed
24

patents filed

PhD, control engineering
Dr.-Ing.

PhD, control engineering

in R&D and software
Since 2008

in R&D and software

narratiQ
Founder & CTO

narratiQ

CONTENTshift 2026
Top-10 Finalist

CONTENTshift 2026

Focus areas

What I do

Agentic AI workflows & LLM applications

Building and operating LLM-driven systems end to end — schema-bound structured output, validation loops, and workflows that call language models without pretending they're autonomous agents.

Python development

Production Python across FastAPI backends, data pipelines, and automation — from a manuscript-analysis platform to internal tooling.

Data science

Data science tooling for lab and R&D work — Python, NumPy, pandas, TensorFlow, scikit-learn — built and adopted by the teams that used it.

Product development as a solo engineer

Full ownership of architecture, backend, frontend, and deployment — twice, as a founder, from problem to production.

Teaching

Teaching graduate students to build LLM systems — including retrieval-augmented generation from scratch — at HS Macromedia since 2023.

Work

Projects

narratiQ

narratiq.de ↗

Founder & CTO — fully remote, sole engineer

An analysis platform that helps publishers review manuscripts. Designed, built, and operated end to end — architecture, backend, frontend, and deployment — as the only engineer. Built as a classical, self-authored Python workflow that calls language models with structured, auditable output, not an agent system: there is no orchestrator and no autonomous agent layer in the product. Runs in production on DigitalOcean and Vercel, currently processing manuscripts in early pilot runs with a publisher. Top-10 finalist, CONTENTshift 2026, with a presence at the Frankfurt Book Fair.

Chapter-aware chunking instead of retrieval

Manuscripts that don't fit a single model call are split along chapter boundaries and refined sequentially — a deliberate alternative to embeddings and vector search, because a full review has to judge the whole work, not just the most similar passages.

Cost telemetry per model call

Every call is logged with prompt, completion, and cache tokens, feeding a daily cost cap — visibility into spend without checking a provider dashboard.

AI stack rebuilt around fewer, declarative calls

Nine separate calls became one merged call plus four genre calls, described as data rather than scattered logic — cost per full analysis dropped to about a third.

Time saved, in my co-founder's estimate

My co-founder, a former publishing director, puts a manual manuscript screening at roughly 25 hours; with the platform it takes about four.

Whispeer

Personal tool — Rust

A small Rust utility that turns speech into text: captures microphone audio, transcribes it through the Whisper API, and places the result directly on the clipboard.

GTD app

Personal tool — Next.js, Supabase

A self-hosted task manager replacing Notion for personal GTD workflows, with an automated processing pipeline for recurring work.

Second Brain & workflow automation

Personal RAG system — Haystack, pgvector, MCP

A production-style retrieval layer that grounds AI coding workflows in a versioned knowledge base. I took hybrid Recall@5 from a reproduced 6/15 baseline to 15/15 through controlled ablations, then achieved 30/30 on a frozen, disjoint holdout set.

Read the engineering case study

HS Macromedia · fully remote

Teaching

Teaching at HS Macromedia since 2023, ten to fifteen students per course. Three Large Language Models seminars taught so far.

  • Software Project 2 — Master
  • Seminar: Large Language Models — Master
  • Seminar: Data Science — Masterupcoming

In the last two Large Language Models seminars, students built retrieval-augmented generation from scratch before reaching for a library: chunking, embeddings from an OpenAI model, cosine similarity, selecting the relevant match, and passing it into the model's context — only afterwards repeating the exercise with the Haystack library.

Deep roots

Research & Patents

Dr.-Ing., Global Chassis Control

TU Dortmund University · 2019

Dissertation on global chassis control of the full vehicle, building on research done at the German Aerospace Center (DLR). Master's-level studies in mechanical engineering with a focus on control engineering, TU Munich, 2003–2008.

Global chassis control research ↗

24 patents

  • 13 · Dispendix — liquid-handling modelling and control, on-site diagnostics, lab process automation
  • 7 · German Aerospace Center (DLR) — vehicle dynamics control
  • 4 · Ocuplus — data preprocessing and output

The 13 patents from the Dispendix work directly enabled that product's market entry.

Timeline

Career

  1. 2025 – today

    Founder & CTO · narratiQ

    Fully remote

    Sole engineer building an AI-powered manuscript analysis platform for publishers, from architecture to production deployment on DigitalOcean and Vercel. Top-10 finalist, CONTENTshift 2026.

  2. 2023 – 2024

    AI & Data Science Expert · Education Partners

    Lippstadt, fully remote

    Automated the migration of unstructured education data into ERP-ready structures, combining Python pipelines with LLM-based extraction from free text. Built an internal AI writing tool adopted company-wide.

  3. 2023 – ongoing

    Lecturer · HS Macromedia

    Fully remote

    Teaching Software Project 2 and the Large Language Models seminar to 10–15 graduate students per course; a Data Science seminar is upcoming.

  4. 2023

    Independent Studies

    Built a broad foundation in LLM systems and automation — expert-chat applications, local and on-prem models, and early agentic workflow patterns.

  5. 2020 – 2023

    Data Scientist, R&D Engineer · Dispendix

    Stuttgart, fully remote

    Solved a technical problem in liquid handling that enabled market entry (13 patents); built data science tooling used across lab, QC, and R&D.

  6. 2020

    Senior Project Manager · HTSG

    Wolfsburg, fully remote

    Led three teams — systems engineering, networking, and a task force — with technical management of 13 employees.

  7. 2014 – 2019

    Head of Software Development, Founder, Managing Director · Ocuplus

    Regensburg, partly remote

    Founded and led the company from product invention (4 patents) to market launch and customer acquisition, hiring and leading two software developers.

  8. 2008 – 2013

    Researcher and Project Leader · German Aerospace Center (DLR)

    Oberpfaffenhofen

    Developed a vehicle dynamics control system (7 patents), the basis of a later PhD, and led DLR's share of a European research project.

Background

About

Johannes trained as a mechanical engineer with a focus on control engineering, then spent five years researching vehicle dynamics control at the German Aerospace Center (DLR) — work that became his PhD. He went on to found and lead Ocuplus, took on data science and technical leadership roles in automotive and medical technology, and — starting at Education Partners in 2023 — taught himself, and since then his students, how to build with large language models. Today he's the founder and sole engineer behind narratiQ, an AI-powered manuscript analysis platform for publishers.

How I work

Everything here — this site included — comes out of a personal, structured way of working with AI: a Second Brain for accumulated knowledge, consistent automation for repeatable work, and agentic development workflows built around Claude Code and Codex. Both are development tools, not runtime components — the distinction matters, and it's one narratiQ's own architecture keeps as well.

Get in touch

Contact

johannes.koeppern@gmail.comLinkedInBook a call