Team Lead · AI
Everest Systems
Everest Systems
Plate IAn atlas of work, in progress· K. Papa · 2026 ·
Solve real problems.
Ship.
Five years of shipping production software. A Master's in AI from the research side. The work I care about turns both into value users can feel.
Now
Based in
Heidelberg
Germany
Germany
Looking for
Real problems
that ship
that ship
II.
Plate II — A portrait of the engineer
An engineer who still writes the code.
Fig. 1 — context/object split,
after the master's thesis
after the master's thesis
I'm an engineer first. I spend most of my days writing the code that ships — not steering meetings around it — even as a team lead. What gets me out of bed is designing, building and owning agentic AI systems that solve real problems for real users, with the right balance of autonomy and human-in-the-loop control.
My background is split intentionally. A Master's at TU Darmstadt taught me to think about intelligence as a research problem: latent spaces, multi-task learning, the structure of representation. Five years at Everest Systems taught me to think about it as a delivery problem: production rollouts, dynamic model routing, tool-use orchestration, every new LLM generation absorbed without rewriting the world.
The through-line is value. I want to point all of this at problems that matter — where careful, well-designed AI compounds human judgment instead of replacing it, and where shipping a feature actually changes someone's day. Health tech, life sciences and biotech are particular obsessions, but the appetite is broader: anywhere the work is hard, useful, and ends in production.
“Hands-on by choice. The work I care about lives in the editor — and ends in something a user can feel.”— On working practice
Fig. 1 / Latent spaceMulti-task · 2022
A schematic — context (blue) and object (ink) latent clusters, separated by a learned boundary.
Especially curious about
Agentic systemsReal-world impactHealth techLife sciencesBiotechRepresentation learningDeveloper tooling
III.
Plate III — Selected work · 2019 → present
Roles, in reverse chronology.
Each entry expands
into its case file.
into its case file.
IV.
Plate IV — Specimen catalog
Selected projects.
Four specimens from
research and production.
research and production.
P/01Specimen
Multimodal Context-Object Split Latent Spaces
Master's Project · TU Darmstadt
A representation-learning framework with separate latent spaces for context and objects, improving diversity in multi-task learning across modalities.
P/02Specimen
Diverse Image Captioning
Research project
Extended deterministic captioning architectures with variational components to trade off caption quality against diversity — bridging computer-vision and NLP techniques.
P/03Specimen
Agentic ERP Workflows
Production · Everest Systems
Multi-agent orchestration over a live ERP. Agents plan, call tools, ask for human approval, and complete real operational tasks. Designed for safe autonomy.
P/04Specimen
Provider-Agnostic AI Backbone
Production · Everest Systems
A model-routing layer with function calling and tool use abstracted over providers. Adopting a new frontier model is a configuration change.
V.
Plate V — Working instruments
Capabilities, distilled.
Field-tested, in production.
Agentic AI & LLMs
Multi-Agent Orchestration (LangGraph)Tool Use / Function CallingModel Context Protocol (MCP)ReAct & PlanningRetrieval-Augmented GenerationHuman-in-the-Loop DesignPrompt Engineering
Languages
Python (Expert)TypeScriptJava
Machine Learning
PyTorchscikit-learnComputer VisionNLP
Engineering
FastAPIReactPostgresVector DBsCI/CD
VI.
Plate VI — Formal education
Formation, in full.
Two degrees, one thesis
on representation learning.
on representation learning.
M
2020 — 2022
M.Sc. Computer Science
Technical University of Darmstadt
Thesis: Multimodal Context-Object Split Latent Spaces for Diverse Multi-Task Learning.
B
2017 — 2020
B.Sc. Business Informatics
University of Mannheim
Thesis (with KSB SE & Co. KGaA): Anomaly detection in time-series data of rotating equipment using machine learning.
Plate VIICorrespondence
Let'sbuild something.
Especially if there's a real problem on the other side of it — bonus points for health, life sciences or biotech, but I'm here for any work that ends in production.