Analytics Engineer - Building reliable data systems close to business decisions.
Before engineering data, I ran a business. Managing inventory, costs, logistics, and margins taught me that data is only valuable when it helps someone make a better decision. That principle guides every system I build.
- Analytics Engineering · Data Pipelines · Data Modeling
- SQL · Python · ETL/ELT
- Data Quality · Testing · API Design
- Production Systems · Business Intelligence
SkyCast - Enterprise climate intelligence platform
Problem: Municipalities lacked unified real-time climate monitoring across 5 data sources.
Built: Modular ingestion pipeline from AEMET OpenData and crowdsourced reports, validated through Pydantic, stored in PostgreSQL with time-series partitioning, served via FastAPI REST API, visualized in Streamlit dashboard with alert engine.
Engineering focus: 158 automated tests in CI/CD, circuit breaker pattern for external API failures, data quality checks at every pipeline stage.
· FastAPI · PostgreSQL · Streamlit · Docker · Python
OptimusPrice - ML-driven hotel revenue management
Problem: Hotels rely on manual pricing or OTAs charging 15-30% commission - slow, reactive, and margin-eroding.
Built: Random Forest regressor with Optuna hyperparameter optimization, feature engineering pipeline extracting 40+ demand signals from booking history and competitor data, served via FastAPI inference endpoint, Streamlit management dashboard with confidence intervals.
Engineering focus: Model interpretability (feature importance scores for stakeholder trust), automated retraining when MAPE exceeds 20%, input validation rejecting out-of-range features.
· Python · scikit-learn · Optuna · Streamlit · Docker
DashLogistics - Multi-market logistics intelligence
Problem: Logistics providers need route-level cost prediction across 50 US states with multiple data sources updating at different cadences.
Built: Multi-source ETL pipeline ingesting fuel prices (AAA API, daily), weather (OpenWeatherMap, hourly), population data, and operational records. XGBoost cost prediction with Optuna tuning. Streamlit dashboard tracking 15+ KPIs with state-level maps. dbt layer with staging, intermediate, and mart models for analytics engineering.
Engineering focus: Data quality gate preventing bad data from reaching the model, decoupled ingestion from prediction (single source failure doesn't block pipeline), confidence scoring for partial feature availability.
· Python · Pandas · XGBoost · Streamlit · PostgreSQL · Docker · dbt
NetTension - EU telecom network stress simulation
Problem: EU regulators need quantitative evidence of the "scissors effect" (traffic growing exponentially while revenue declines) to inform infrastructure policy.
Built: DuckDB-powered analytical pipeline processing 41,937 rows of CNMC regulatory data joined with 1.8M+ Eurostat macroindicators. Six pre-registered hypotheses tested using regression analysis and time-series decomposition. Power BI dashboard for interactive exploration.
Engineering focus: Hypothesis-first methodology preventing confirmation bias, DuckDB enabling sub-30-second query times on a laptop without infrastructure provisioning, fully containerized reproducible analysis.
· Python · DuckDB · Power BI · Docker
Full case studies with architecture diagrams at juandelaf1.github.io
- Golden Forecast - ML forecasting pipeline and dashboard for gold price prediction. End-to-end: extraction, 35 engineered features, 12 models, Docker deployment with CI/CD.
- dbt on DashLogistics - Analytics engineering layer: staging, intermediate, and mart models transforming logistics data for dashboard consumption.
- Understand - Map the business context, identify the decision, define success.
- Model - Design architecture before writing production code.
- Build - Implement with tests at every layer, validation at every boundary.
- Measure - Validate output against the original business problem.
Some repositories have been archived or set to private to reflect my current Analytics Engineering focus:
- Private: academic projects and early prototypes
- Archived: mintlify-docs, docs, DataScope (fork), and early Python scripts
- Public featured repos: SkyCast, OptimusPrice, DashLogistics, NetTension, Golden Forecast
- Public experiments: EnRuta, Attention Observatory, Spanish Wine EDA