I turn messy data into clear decisions.

Quang Dat LE is a data scientist and software engineer based in Paris, France. He is a data scientist at BNP Paribas and previously worked as an R&D engineer at Valeo. He builds models, pipelines and tools that reach production, from LLM document extraction to machine learning written from scratch. This page lists his data and machine-learning projects; the main site covers the websites, business tools and automation he builds for companies.

Experience

BIFORA — Data Scientist, 2024 to 2025. Migrated R, SAS, VBA and SQL code to Python for major accounts; built and deployed machine-learning pipelines in Dataiku with Power BI dashboards; mentored a junior, doubling their output. Result: 30% less data-processing time, twice the junior output. Stack: Python, Dataiku, Power BI, SQL.
Valeo — R&D Engineer, 2023 to 2024. Set up code-quality gates with SonarQube and test-driven development; ran DevOps pipelines with Jenkins and Docker; built multiplatform tools in Java, C++ and Python plus encryption for critical systems. Result: 50% fewer deployment errors, 25% more system efficiency. Stack: Java, C++, Python, Jenkins, Docker, SonarQube.
Phan&LeNoble — Drupal Developer, 2022 to 2023. Built object-oriented PHP modules and ERP features on Drupal, designed optimized databases with UML modeling, and delivered modern interfaces. Stack: PHP, Drupal, JavaScript, UML.

Selected projects

Paris House Pricing (2026) — Real-estate intelligence over roughly 1,300 Paris neighbourhoods: free open data scored 0 to 100, a LightGBM price-growth forecast with SHAP and a walk-forward backtest, served on a live map. Stack: Python, LightGBM, PostGIS, Airflow, MapLibre, SHAP.
Monte Carlo Stock Research (2026) — A stock-research API streaming GARCH, jump and ensemble Monte Carlo price paths, options flow (GEX), a breakout scanner and backtests over WebSocket. Stack: FastAPI, GARCH, WebSocket, Alpaca.
Home Loan Scoring (2026) — A credit-risk pipeline from raw data to a score_customer() function: an XGBoost approval model with SHAP explainability and clear risk bands. Stack: XGBoost, scikit-learn, SHAP.
Sector Lagging Scan (2026) — Finds stocks lagging their sector with a stationarity-correct method: regress log price on the sector, ADF-test the residual spread, z-score it, and flag laggards likely to revert. Stack: Python, statsmodels, yfinance.

Also built

A World Cup 2026 predictor (XGBoost, Dixon-Coles, Monte Carlo, Streamlit, 2026), LLM tax-document extraction (OCR, LLaMA 3.1, Mistral, 2025), an MNIST API written from scratch (Flask, pure Python, 2023), a cephalometric measurement tool (OpenCV, Tkinter, 2024), a Nutri-Score analysis pipeline (scikit-learn, KNN, PCA, 2025), DataScope for exploratory analysis (pandas, plotly, seaborn, 2025), and AI video translation (Whisper, FFmpeg, 2024).

Stack

Data and ML: Python, scikit-learn, Pandas, NumPy, Dataiku, Power BI, Sparrow
AI and generative AI: LLaMA 3.1, Mistral, DeepSeek, Whisper, named-entity recognition
Engineering: Java, C++, C, JavaScript, Angular, Flask, Laravel
DevOps and QA: Docker, Jenkins, Git, SonarQube, PyTest, CI/CD
Data engineering: SQL, PostgreSQL, PostGIS, R, SAS, VBA

Contact

Email lequangdat1071@gmail.com, GitHub github.com/qle107, LinkedIn linkedin.com/in/qdat. Based in Paris, France. Replies within a day.