Applied
Engineering
Results
samuel-torres.com
Role
Building a small end-to-end ETL pipeline: pull crypto price data from a public API, transform it into an analysis-ready shape, and store it efficiently for later querying — a personal project.
Core stack
Python · pandas · Parquet · requests · ETL · Data Engineering
Code
What it is
A small, complete ETL (Extract, Transform, Load) pipeline built to practice data engineering fundamentals on real, freely available market data. It pulls cryptocurrency price data from a public API, cleans and reshapes it with pandas, and writes the result to a columnar format built for repeated analysis rather than one-off reads.
How it's built
- Extraction: scheduled requests against a public cryptocurrency price API using Python's
requestslibrary. - Transformation: cleaning, type conversion and time-series resampling/aggregation with pandas.
- Load: writing the transformed data to Parquet — a columnar, typed, compressed format — instead of CSV.
- Extract, transform and load kept as separate steps/functions, so each stage can be tested, rerun or replaced independently.
Design notes
Parquet stores typed, columnar data with compression built in — reading a handful of columns out of a large time series is far cheaper than parsing an entire CSV file every time.
Re-running the pipeline for a date range it already processed shouldn't create duplicate rows — the load step is written to check what's already there before writing.
Extraction doesn't know about pandas, and the transform step doesn't know where the data came from or where it's going — each stage can be tested and reasoned about on its own.
Short version (portfolio card)
An ETL pipeline pulling cryptocurrency price data from a public API, transforming it with pandas, and loading it into Parquet for efficient repeated analysis.
Resume / LinkedIn bullets
- Built an ETL pipeline in Python extracting cryptocurrency price data from a public API on a schedule.
- Transformed time-series price data with pandas (cleaning, type conversion, resampling).
- Loaded processed data into Parquet, choosing it over CSV for columnar compression and typed schema.