Paul Gradischnig

KELAG · Lakeside Park, Klagenfurt

• Four weeks on one question: what does a smart meter with 15-minute readings reveal about a household? More than expected. Photovoltaics, battery storage, EV charging, electric boilers and electric heating each leave their own signature in the load profile.
• Trained models on several thousand households from open research datasets, with feature engineering over the raw time series.
• Built an interface that evaluates a single load profile against those models.

• Analyzed 15-minute smart-meter load profiles to detect photovoltaics, battery storage, EV charging, electric boilers and electric heating.
• Trained models on several thousand households from open research datasets with time-series feature engineering; built an interface that evaluates a single load profile.

KELAG · Carinthia

• Independently built and tested a Chaos Toolkit extension for Azure Container Apps to improve cloud observability and system resilience.
• Designed a sidecar approach to inject CPU/memory stress into serverless containers without host access, the core of my diploma thesis.
• Shipped infrastructure as code with Azure Bicep and CI/CD via GitHub Actions.

• Built and tested a Chaos Toolkit extension for Azure Container Apps to improve observability and resilience (diploma thesis).
• Designed a sidecar that injects CPU/memory stress into serverless containers without host access; IaC with Azure Bicep, CI/CD via GitHub Actions.

Kostmann GesmbH · Carinthia

• Digitalized business processes, maintained data, and built/analyzed reporting.

• Digitalized business processes, maintained data, built reporting.

Infineon Technologies · Villach

• Supported hardware maintenance in cleanroom production and assisted the IT team with reporting.

• Hardware maintenance in cleanroom production; IT team reporting.

• Platform that aggregates digital grocery flyers and finds the cheapest store for a shopping list. Built with Arbeiterkammer Kärnten, funded by Innovation@School.
• Work on the backend architecture and AI pipeline: flyer parsing, product normalization, price history. Part of a three-person team.

• Platform that finds the cheapest store for a grocery list from digital flyers; with Arbeiterkammer Kärnten, funded by Innovation@School.
• Backend architecture and AI pipeline (flyer parsing, normalization, price history), in a three-person team.

• Proxmox cluster whose third vote runs off-site: a QDevice at Hetzner, 600 km away. Plus self-hosted AI (embeddings + vector DB, OpenCLAW), GPU sharing, and Wake-on-LAN over a WireGuard tunnel.

• Proxmox cluster with its third vote off-site (QDevice at Hetzner, 600 km away); self-hosted AI, GPU sharing, Wake-on-LAN over WireGuard.