Turning financial complexity and luxury market signals into predictive intelligence
Data Scientist with a dual specialization in quantitative finance and luxury market analytics. I build end-to-end pipelines — from raw market data ingestion to interpretable ML models — with a strong emphasis on statistical rigor, reproducible research, and production-ready deployment.
- Finance track: NLP on Fed communications, econometric panel models, portfolio optimization
- Luxury track: demand forecasting, pricing analytics, consumer sentiment modeling
- Stack: Python · R · SQL · PyTorch / TensorFlow · Qiskit · Streamlit · Docker
Currently exploring quantum machine learning for financial optimization and LLM-augmented research workflows.
Languages
Machine Learning & Deep Learning
Data & Visualization
Infrastructure & MLOps
Quantum Computing
| Project | Domain | Stack | Status |
|---|---|---|---|
| sentiment-powell-nlp | Fed communication analysis · NLP | Python · BERT · HuggingFace · TensorFlow |  |
| panel-project | EU GDP determinants · Econometrics | Stata · Python · Streamlit | |
| Quantum_Computing | Quantum ML · Portfolio optimization | Python · Qiskit · PyTorch |  |
| academic-stress | Behavioral data analysis | Python · R · Statistics |  |
| linux-sys-monitor | System observability | Python · Linux · Shell | |
Problem: Quantify the hawkish/dovish stance of FOMC press conferences over time to anticipate monetary policy shifts.
Approach:
- Transcript collection and preprocessing (tokenization, TF-IDF, stopword removal)
- Fine-tuning a BERT model on financial sentiment corpora (Stanford Sentiment Treebank + custom labels)
- Embedding layers (dim=100) → BiLSTM / Deep CNN → sentiment score per conference
- Statistical validation: Wilcoxon signed-rank tests with Bonferroni correction
- Time-series overlay with rate decision outcomes
Key finding: Dovish language clusters (identified via Bag-of-Words) show a statistically significant lead of 2–3 sessions before rate cuts (p < 0.01).
- Building: Interactive Streamlit dashboard on top of
panel-projecteconometric models - Learning: Quantum kernel methods with Qiskit for near-term quantum devices (NISQ)
- Reading: Advances in Financial Machine Learning — Marcos López de Prado
- Open to: Research collaborations at the intersection of NLP, quantitative finance, and luxury market analytics
All my projects follow a reproducible research structure:
project/
├── README.md # Executive summary, results, install guide
├── notebooks/ # Numbered EDA & modelling iterations
├── src/ # Production-grade, modular Python code
├── data/ # Anonymised samples only (.gitignore for raw data)
├── report/ # Final PDFs, high-res figures
├── dashboard/ # Streamlit / Dash app + Dockerfile
├── requirements.txt # Pinned dependencies
└── LICENSE
"Data without context is noise. Context without data is opinion."