Vottas
Data Scientist/AI Engineer specialising in NLP and Gen AI. Deployed Greece's first local zero-cost LLM in a IBM's CP4D. Published researcher at NCSR Demokritos. Currently pursuing a full-time MBA at the University of Piraeus — elected Vice-President of the class.
- Deployed Greece's first local zero-cost Llama-based LLM inside IBM's Cloud Pak for Data (CP4D) — a first-of-its-kind achievement in Greek banking. Allowed experimentation, development without cost and was the first step towards our team's focus on GenAI applications.
- Designed and implemented an automated AI reporting system delivering weekly customer complaint insights to C-Level executives and quarterly reports to the Executive Committee (ExCo).
- Developed 5 production-grade models currently used. Models used for sentiment analysis both with Neural Networks and GenAI, text summarization using LLMs, test classification utilizing a combination of Neural Networks, LLMs and business rules.
- Migrated, automated and currently maintaining 20+ production-level ML models on cloud infrastructure.
- Currently focusing in development of AI Agents using Microsoft's Copilot Studio.
- Published peer-reviewed research on drug-drug interaction (DDI) prediction for lung cancer at IEEE CBMS 2025.
- Chosen as one of the best papers and called for publishing a special issue version.
- Built a disease-specific biomedical knowledge graph (375,707 nodes, 17.9M relationships) from PubMed and MEDLINE.
- Developed a path-analysis ML approach (BLGPA) outperforming all graph embedding baselines (TransE, HoLE, RESCAL, DistMult).
- Assisted in the development of the Data Warehouse for Credia Bank (formerly Attica Bank), and for GasLog (shipping company).
- Developed a PowerBI report for water monitoring at Metlen's White Houses.
- Helpdesk and sysadmin duties for a military Brigade and Military Hospital.
- Trained doctors and nurses on a new clinical software system during nationwide healthcare digital transformation rollout.
Deep learning system for automated medical image diagnosis. Classifies brain MRIs for tumour detection and chest X-rays for pneumonia screening. Built with convolutional neural networks to assist clinical decision-making.
End-to-end system for predicting drug-drug interactions using a biomedical literature knowledge graph. Outperforms graph embedding baselines. Includes a real-world list of novel DDI predictions not yet documented in DrugBank.
Deep Learning · ML · NoSQL/SQL · AI Ethics · XAI
Python · ElasticSearch · Data Mining · Data Structures · C
Constructs a lung-cancer-specific biomedical knowledge graph (375K nodes, 17.9M relationships) from PubMed/MEDLINE to predict five classes of drug-drug interactions. A path-analysis ML approach (BLGPA) outperforms TransE, HoLE, RESCAL and DistMult graph embeddings. Funded by EU Horizon 2020 SIMPATHIC project.