Stylianos I.-Engineering Consultant and Broker

• Architecting an end-to-end AI platform for de novo drug discovery and systems pharmacology, specifically targeting novel neuro-therapeutics. Formulated a comprehensive, investor-ready business model integrating automated intellectual property strategy and novel risk-hedging mechanisms.
• Engineered deep reinforcement learning (DRL) pipelines to guide molecular generation. Developed custom, multi-objective reward functions to iteratively optimize compounds for synthetic accessibility (SA), quantitative estimate of drug-likeness (QED), and receptor binding affinity.
• Designed and deployed complex retrieval-augmented generation (Graph RAG) architectures powered by LLM committees. Utilized these multi-agent workflows to autonomously parse biochemical literature, identify therapeutic targets, and map pharmacological mechanisms.
• Built graph neural networks (GNNs) for advanced property prediction, leveraging generative models to map novel chemical spaces and heavily accelerate in silico lead optimization cycles.
• Developed a custom, open-source kdb+-inspired analytical database in Julia to handle high-velocity data processing. Engineered the core vector storage layer utilizing PostgreSQL with pgvector for efficient semantic search.
• Managed the entire computational environment, optimizing deep learning workloads for parallel execution using heavy CUDA/GPU acceleration within a Dockerized Ubuntu Linux (WSL) environment.

Technologies, Libraries & Frameworks Used: PyTorch, Graph Neural Networks (GNNs), Hugging Face Transformers, Ray RLlib, Stable Baselines3, DeepChem, RDKit, OpenBabel, LangChain, LlamaIndex, OpenAI/Anthropic APIs, Julia, PostgreSQL (pgvector), Docker, CUDA, Ubuntu Linux (WSL), Google Compute Engine

• Currently developing an open-source, high-performance database engine in Julia, engineered as a modern, AI-native analogue to the kdb+ ecosystem. Driven by a deep architectural analysis of decompiled kdb+ structures, the project replicates its ultra-low latency, in-memory column-store mechanics while natively integrating vector storage capabilities. By fully leveraging Julia’s JIT compilation and multiple dispatch, this system is designed to handle high-frequency continuous data ingestion and complex multi-dimensional similarity searches. The engine serves as a highly optimized infrastructure layer, providing enterprise-grade speed for quantitative modeling and large-scale machine learning pipelines without the overhead of proprietary licensing

• Participated in TERMINET and PHOENIX EU research projects
• Developed ML classifiers for an industrial-protocol focused Security Information and Event Management (SIEM) based on SPEAR EU project research

• Full stack web development in Node.js and Python
• Worked on Google Clusterfuzz codebase and its ML modules
• Developed a new startup web project

• Investment fund focused mainly on cryptocurrencies and Software Engineering
• Conducted algorithmic trading for cryptocurrencies, ETFs and stocks with Python ML models (XGBoost, Zipline, Keras, Pandas)
• Performed cryptocurrency backend programming in C#, Erlang and Python
• Developed experimental privacy-focused blockchain for prediction markets in Erlang
• Created deep learning (LSTM), random forest and boosting models for trading signal generation
• Performed blockchain graph analysis for transaction tracing using MongoDB, Cassandra and Python

• Compiled study on socioeconomic and environmental impact of using Port of Larnaca as logistics base for offshore exploration drilling operations in Cyprus

• Conducted feasibility studies on LNG ship fuel use in European ports for COSTA project
• Modeled freight rates and optimized cement carrier fleet deployment for TEN-T Executive Agency project

• Designed mechanical installations for industrial buildings
• Issued installation and operation permits for factories
• Compiled environmental impact assessment studies for small and medium enterprises