Hello, I'm
Juan Daniel
Hernandez Vargas
Data Scientist & Specialization Candidate
Statistical modeling, machine learning and trustworthy data that turn complex information into clear decisions.
Hello, I'm
Data Scientist & Specialization Candidate
Statistical modeling, machine learning and trustworthy data that turn complex information into clear decisions.
I am a Data Scientist currently completing a professional specialization in Statistical Analytics. My background in Biology gave me a scientific way of working: frame the question well, check the evidence, and justify every decision.
I work with data from different domains, from credit risk, fraud detection and customer retention to urban mobility, with one common thread: understand the data before modeling it, choose methods that fit the problem, and build explainable results that can be turned into decisions.
Looking ahead, I want to bring that same rigor to biological databases, where my scientific training and my statistical toolkit meet.
Specialized in probability, regression, and hypothesis testing.
Predictive classification and regression models with rigorous validation.
Validation rules, anomaly detection and imputation grounded in domain knowledge.
Interpretable models with SHAP, so every result can be understood and audited.
Data Scientist specializing in Statistical Analytics, Machine Learning and data quality for decision-making.
Drag or scroll to explore →
IEEE-CIS (Vesta) — 590,540 transactions
End-to-end project: SQLite builds each card's history with window functions, a LightGBM with chronological validation ranks the alerts, and a PyTorch autoencoder looks for fraud without labels. Reviewing the riskiest 1% of transactions, 85% of alerts are actual fraud.
NYC Taxi & Limousine Commission — 9.7M trips
Interactive dashboard that maps demand by zone, day and hour and audits every fare against the official rate card. It showed that peak-demand hours yield the lowest revenue per occupied hour and that fare inconsistencies cluster by technology vendor, not by driver.
Banco de la República — 416 months (1991–2026)
Dashboard that turns an ARIMA + GARCH model into the probability that the dollar exceeds a reference rate in each month, useful for budgeting. It showed that the point forecast does not beat a random walk, but the GARCH bands are well calibrated: 6.7% of months fall outside the 95% band, versus 11.7% with constant variance.
Technical assessment — financial institution
Full credit scoring pipeline with OOT validation, SHAP audit, and ethical AI analysis. Compared Gender-inclusive vs Gender-blind models — a 1.96 pp GINI loss for a fairer system. Neither external vendor outperformed the internal model.
Personal Project
One click logs a job posting into a Google Sheet: a Chrome extension scrapes the listing, an Apps Script backend drafts the outreach message and a cover-letter PDF with Claude, and a scheduled Gmail job advances each application's stage on its own. No API key ever ships inside the extension — every credential stays server-side.
Beta Bank
Identified high-risk segments — older, inactive clients in Germany — enabling targeted retention campaigns. Random Forest with minority-class upsampling delivered the best balance of precision and recall.
I am currently open to new opportunities in Data Science and Data Analytics across industries. Whether you have a specific inquiry or just want to connect, I'd love to hear from you.