Marcel Huber

Co-founder, Woodquarter Labs

Austria

• Built Bowl Builder, a browser app for parametric 3D-printable bowls with Voronoi lattices and support-free overhangs.
• Built LanternStudio, a browser configurator for modular 3D-printable lanterns: over 75,000 frame combinations, a wall editor with image tracing, and print-ready Bambu Studio exports.
• Built the platform around Bowl Builder: accounts, a design gallery, tiers and access codes, deployed through CI.
• Wrote printviz, a Blender pipeline that renders the sliced print layer by layer for product videos.
• Ran the MakerWorld crowdfunding launch, including the listings and launch videos.

• product
• TypeScript
• Three.js
• manifold-3d
• Blender
• Next.js

Gen AI Data Scientist, Mimirio

Linz / remote

Seven-person startup building an agentic AI platform for organisations with sensitive data. It answers from internal documents and cites the exact source for every claim. Customers include Raiffeisen Informatik.

• Owned citation handling in the production RAG agent: traced and fixed inconsistencies in how citations were created, tracked and passed from backend to frontend.
• Built 5+ agent tools that shipped to customers including Raiffeisen Informatik: web search, full-document reading, email drafting, Jira ticket handling, and mentioning a specific document or ticket by name.
• Connected the agent to Microsoft SharePoint and Jira, including automatic draft replies when a new ticket arrives.
• Found a context-management bug that sent the same metadata to the model twice; the fix cut response time by a few seconds per message, depending on its length.
• Built three synthetic benchmark datasets and measured every prompt and pipeline change with them, alongside unit tests, integration tests and traces. The CTO praised my lab reports on these experiments for their clarity and thoroughness.

• RAG
• agents
• agentic loop
• tool calling
• Haystack
• Claude
• synthetic data
• evaluation

“We would rehire Marcel at any time and without reservation.”

Student Researcher, Medical University of Vienna

Vienna

• Adapted NVIDIA's MAISI framework to 2D OCT and built the pipeline from scratch: intensity handling, four disease classes, and patient-level splits to prevent data leakage.
• Trained a 256 px adversarial VAE (SSIM 0.892, reconstruction FID 5.00), a class-conditional latent diffusion model with rectified flow, and mask-conditioned models on RETOUCH.
• Showed that synthetic scans help classifiers when real data is scarce: +0.071 mean Macro F1 at 10% real data, across ten seeds on a patient-disjoint test set.
• Compared four counterfactual editing approaches and showed that localised edits work, while classifier scores alone cannot confirm that a change is medically meaningful.

• latent diffusion
• rectified flow
• VAE
• MAISI
• OCT
• synthetic data
• PyTorch

Intern, WEB Windenergie AG

Pfaffenschlag

• Automated a colleague's recurring tasks, saving half a workday per week.
• Wrote Python scripts for data fetching and transformation.
• Upgraded the Raspberry Pi infrastructure and PostgreSQL schema; added backups and documentation.
• Analysed the efficiency of a four-year research project.

• Python
• PostgreSQL
• Raspberry Pi
• automation

Fellow, Linz Institute of Technology

Linz

• Designed the SQL schema from entity-relationship diagrams.
• Built a recommender using BERT embeddings for non-numerical data.
• Used PCA and t-SNE on the latent embeddings to improve recommendations.
• Explored converting PDFs to EU-accessibility-compliant formats with an LLM.

• BERT
• recommender systems
• SQL
• t-SNE

MSc Artificial Intelligence

Johannes Kepler University Linz, final exam pending, current average 1.6 (Austrian scale), GPA 3.40 / 4.00

Thesis Generating Counterfactual OCT Retinal Images: Training Generative AI Models (Diffusion) for Class-Specific Image Transformation

BSc Artificial Intelligence

Johannes Kepler University Linz, grade 1.9 (Austrian scale), GPA 3.10 / 4.00

Thesis Generative models, synthetic data and counterfactuals: applications in retinal imaging

MARIO challenge finalist, MICCAI 2024

Methods for diagnosing and monitoring age-related macular degeneration from OCT. Presented in Marrakech; the paper was published by Springer.

1st place, Danube Dynamics image segmentation challenge

Winning entry in an image segmentation competition.

3rd place, AI Cloudflight Coding Contest

Timed AI coding contest at JKU.