EAF Endpoint Control
Hybrid physics + ML steel temperature and chemistry prediction with closed-loop recommendations — part of a program delivering up to ~20% tap-to-tap time reduction and ~8% energy savings at a European steel plant.
# try asking:
# or type a command: help · ls projects/ · theme amber
I’m a senior data scientist in Trier, Germany, with a background in applied mathematics and 4+ years of experience shipping production-grade AI in heavy industry. My specialty is hybrid physics + machine learning: combining first-principles process models with learned components — most notably closed-loop endpoint control for large-scale electric arc furnaces.
I work across the full stack, from OPC UA / Siemens S7 signal ingestion and InfluxDB / PostgreSQL pipelines to ONNX model deployment, Grafana operator cockpits, and fully on-premise LLM agents orchestrated via MCP. Beyond industry, I’m exploring quantitative finance, causal inference, and privacy-focused local AI.
Hybrid physics + ML steel temperature and chemistry prediction with closed-loop recommendations — part of a program delivering up to ~20% tap-to-tap time reduction and ~8% energy savings at a European steel plant.
Zero-training probabilistic forecasting module returning p10/p50/p90 quantile bands on any plant signal. Live on 12+ EAF cooling panels as a first-class Industream product.
Fully on-premise LLM advisory layer on top of hybrid models: explains risk and proposes countermeasures to operators, grounded in live signals via MCP — no data leaves the plant.
Reusable no-code ML components for time-series, image and sound data in an industrial AIoT platform — cutting time-to-deploy for new use cases from weeks to hours, packaged ONNX-first.
This site: a dependency-free terminal UI with a streaming AI assistant that has real tools — page navigation, CV download, and taking messages for me. See how it works.
full breakdown: type cat skills.txt in the terminal above.
End-to-end ownership of production AI: hybrid physics + ML endpoint control for electric arc furnaces, probabilistic forecasting (Forecastream), causal ML, Grafana operator cockpits, and a fully on-premise LLM Plant Advisor orchestrated via MCP. Mentoring junior data scientists on production engineering.
Four generations of deep learning models for closed-loop EAF voltage control: ~10% more power-on time and ~2 extra heats per day across two plant shells — a seven-figure annual margin uplift per furnace. Official SMS group reference letter for exemplary performance.
Predictive maintenance in steel: ML/DL time-series models for anode temperature prediction, data validation pipelines, and anomaly detection on noisy industrial data.
Foundation in applied mathematics, machine learning, and data science.
This site is a dependency-free static page with a real AI agent behind it — built to stay simple, cheap, and hard to abuse:
browser ──POST /api/chat──▶ Cloudflare Pages Function ──▶ OpenAI Responses API
▲ │
│ ├─ knowledge: /data/profile.json (server-side)
└──── NDJSON stream ◀──────────┤
text deltas + UI actions └─ tools: show_section · offer_cv · leave_message
help, ls projects/,
theme amber and friends run entirely in your browser — zero API cost.