I ask the right questions to uncover the most impactful insights. I build end-to-end data projects: from extracting and cleaning real-world data to analysis, visualization, and automation. Certified in Google Advanced Data Analytics and Google Data Analytics.
Public Spending Analysis — Chile's Public Market (Mercado Público)
PythonpandasNumPyMatplotlibSeabornPower BI
Interactive Power BI dashboard: total spending, top 10 procurement agencies, regional map, and category filters that recalculate all visuals in real time.
JUNAEB and CENABAST together concentrate ~30% of total spending. The curve flattens quickly: a few large agencies drive most of the spending, leaving a long tail of minor buyers.
The Metropolitan Region (RM) accounts for ~62% of the total budget, and the top 5 regions capture ~81%. Geographic concentration is significantly stronger than sector-based concentration.
Spending across industries is highly distributed: the 80% cumulative mark is only reached near the top 14 categories, indicating a broad and open market.
No single category dominates: the cumulative 80% threshold is not met even within the top 15 categories, signaling a highly diverse market with multiple entry points.
An EDA of ~1.1 million public purchase orders (ChileCompra) to answer where businesses should position themselves as government suppliers. Contraintuitive finding: a low supplier count does not guarantee a business opportunity—true niches lie in high-spending markets with weak leaders, rather than in closed, captive markets.
The risk paradox: while Grade G is the most dangerous on an individual loan level (49.4% default rate), it only accounts for 1.8% of total portfolio defaults. Three grades (C/D/B) concentrate 73.8% due to their massive volume.
The credit portfolio deteriorated as origination volume grew: defaults spiked upwards in 2014, reaching a peak in 2016 (23.7%). The apparent "improvement" in 2018 is an incomplete cohort artifact, explicitly flagged as unreliable.
When crossing grade × purpose, the grade heavily dominates: the most toxic combination is Grade G + debt consolidation (50.6%). High-volume, high-risk pockets are prime candidates for pricing premium or rejection policies.
Credit risk analysis conducted on 1.23 million peer-to-peer loans leveraging advanced SQL (window functions, CTEs). Key finding: three grades (C/D/B) concentrate 73.8% of all defaults due to their sheer volume, not just their individual rates—proving that risk monitoring must prioritize volume × rate over the default rate alone.
Inflation ETL Pipeline — Central Bank of Chile (BCCh)
PythonrequestspandasSQLAlchemyPostgreSQLPower BI
An automated, end-to-end data pipeline that extracts UF and CPI series directly from the Central Bank API, transforms the data, loads it into PostgreSQL using idempotent writes, and visualizes it in Power BI. It runs completely unattended once a month. Finding: the Chilean Peso lost ~38% of its purchasing power over 11 years, while savings held in UF preserved their value almost perfectly.
