I build retrieval-augmented generation systems that actually work in production. My master's thesis — a hybrid GraphRAG chatbot serving ~880 scientific papers — combined a Neo4j knowledge graph, dense vector search, and multi-query reformulation into a deployed application (algaebot.filipnovy.dk). A paper isolating the contribution of each pipeline component is under review at KONVENS 2026.
Previously I spent 1.5 years at T-Mobile Czech Republic engineering SQL data pipelines and shipping Power BI dashboards adopted by regional sales teams. I'm looking for AI/ML engineering roles where rigorous evaluation matters — not just building demos, but measuring whether they work.
Python (PyTorch, LangChain, FastAPI, Pydantic), SQL, R, Java basics
Transformers, dense embeddings, cross-encoder reranking, RAG / GraphRAG, vector DBs (ChromaDB), knowledge graphs (Neo4j)
OpenAI / DeepSeek APIs, prompt engineering, RAGAS, custom eval frameworks, structured output (instructor), agentic workflows
Git, Docker, Streamlit, Power BI, Plotly, DigitalOcean
Production RAG system over 879 peer-reviewed PDFs. Full pipeline: PDF extraction with OCR fallback, recursive chunking (~32k chunks), knowledge graph construction (~60k entities, 100k+ edges), two-stage summarization, multi-query retrieval with cross-encoder reranking. Factorial ablation study isolating graph expansion and community summaries with statistical validation. +4.8% faithfulness over baseline. Paper under review at KONVENS 2026.
From-scratch implementations: CNNs, RNNs/LSTMs, autoencoders, VAEs, GANs, Transformers, plus backpropagation and SGD variants. Built across coursework and self-study.
Fine-tuned ViT with custom attention heads. Highest grade (12/12) on oral exam.
End-to-end trading advisor: Weibull survival model for order-fill probability, linear program for capital allocation. Ranked top 0.1% of 460,000+ accounts by actively traded capital.
Engineered SQL pipelines against Oracle and MS SQL to consolidate revenue data. Designed Power BI dashboards adopted by regional sales teams across Czech Republic. Translated between business stakeholders and analytics team.
Mentored incoming MSc students across 30+ nationalities. Recognized in written recommendation by Programme Coordinator.
Thesis: Domain-Specific Chatbot for Algae Research — A Hybrid GraphRAG Approach. Supervisor: Tariq Yousef. Coursework (grade 12): Deep Learning, Applied ML, NLP, Data Visualization.
Thesis: An Analysis of the Current State of Research on Artificial General Intelligence (grade A). Erasmus exchange in Poland.
Controlled factorial ablation evaluating retrieval components over 879 scientific papers. Identified super-additive interaction between graph expansion and community summaries. Optimized configuration achieves +4.8% faithfulness.
