Data Scientist at INGKA Group (IKEA)
🏆 Oral Presentation at CLeaR 2025: My paper on Combining Causal Models for Accurate NN Abstractions was selected for an oral presentation. I study how models reason, not just what they predict.
- Languages: Python, Golang, SQL
- Machine Learning: PyTorch, TensorFlow, Scikit-learn
- LLMs & GenAI: Hugging Face (Transformers, PEFT), LangChain, OpenAI API
- Causal Inference: DoWhy, CausalML
- Infrastructure: Google Cloud Platform (Vertex AI, BigQuery)
-
Combining Causal Models for Accurate NN Abstractions Accepted for Oral Presentation at CLeaR 2025. Investigating the mechanistic interpretability of GPT-2 through causal abstractions.
-
Reproducibility Study: Label-Free Explainability Published reproducibility study on unsupervised model explainability.
-
Cognitive Swarm: Real-time Semantic Attention Mapping A "Vibe-Coded" AI Agent for focus. Built a Chrome Extension + Web App that tracks user browsing through 384-dim vector embeddings (
all-MiniLM-L6-v2).- The "North Star" Metaphor: Maps real-time attention drift in a polar-coordinate "constellation map."
- Tech: Supabase Edge Functions, pgvector similarity search, D3.js visualization, and Framer Motion for interstellar UI physics.
-
Romanian Political Discourse Analysis In Progress. Analyzing the discourse of politicians in official meetings using NLP techniques.
