CV

Energy Consultant

WMG Sustainable Innovation

• Client-facing delivery on deep-tech and EU-funded programmes in hydrogen technologies, manufacturing data spaces, and shipyard digitalisation.
• Proposal development, technical scoping, and meeting preparation with industrial and research partners.

MSc

Technical University of Denmark (DTU)

Engineering – Physics & Nanotechnology

• Thesis: optimisation of electrodeposition parameters for hydrogen evolution reaction (HER) catalysts.
• Specialisation: electrochemistry and green hydrogen.

BSc

National and Kapodistrian University of Athens (NKUA)

Physics

catalyst-kg-agent

Knowledge-graph-grounded, cost-aware multi-agent system for materials-discovery campaigns.

• Decides between a knowledge-graph lookup, an MLIP surrogate query (MACE), and an expensive simulated experiment under an explicit budget, as in a self-driving-lab workflow.
• Materials Project knowledge graph of 189 HER/OER-relevant materials; zero-shot MACE reaches 0.114 eV/atom MAE on non-oxides.
• CGCNN baseline evaluated with composition-disjoint cross-validation to control polymorph leakage.
• Stack: PyTorch Geometric, MACE, BoTorch/Ax, pydantic-ai, Ollama, NetworkX, MLflow.

notion-second-brain

Fully local retrieval-augmented agent over a personal Notion workspace.

• Hybrid dense + BM25 retrieval with reciprocal-rank fusion and cross-encoder reranking.
• Persistent memory, OCR ingestion of PDFs and images, and an anchored-rubric evaluation harness.
• Stack: pydantic-ai, Ollama, Qdrant, Streamlit, Docker.

CaMEL-RAG — LLM Hackathon for Materials Science & Chemistry

Retrieval-augmented LLM for natural-language queries over Open Catalyst adsorption-energy data (team Code4Catalysis-KFUPM).

• FAISS vector store over sentence-transformer embeddings; answers cite the retrieved records they are grounded in.
• Returned stored DFT adsorption energies exactly on 500 in-index test queries (R² = 1.00), validating retrieval and grounding.
• Stack: Python, FAISS, sentence-transformers, OpenAI API.

BayBE one more time — Bayesian Optimization Hackathon for Chemistry and Materials

Bayesian optimisation with BayBE to screen small-molecule corrosion inhibitors for aluminium alloys (team Surface Science Syndicate).

• Compared one-hot, Mordred, RDKit and Morgan-fingerprint encodings against random search in simulated 50-experiment campaigns (10 Monte Carlo repeats).
• Transfer learning from AA1000 data reached ~96% inhibition efficiency on AA2024 by the 12th experiment, versus ~89% after 25 without it.
• Stack: BayBE, RDKit, Mordred, pandas.

From Knowledge to Action: Outcomes of the 2025 Large Language Model (LLM) Hackathon for Applications in Materials Science and Chemistry

arXiv

Community paper; contributed CaMEL-RAG (team Code4Catalysis-KFUPM).

Bayesian Optimization Hackathon for Chemistry and Materials

ChemRxiv

Community paper; contributed BayBE corrosion-inhibitor study (team Surface Science Syndicate).