Note
A data platform for a three-person data team: Prefect, DuckDB and AWS
Python and Prefect pipelines into a DuckDB-over-Parquet warehouse on AWS, serving embedded Power BI to 45 internal and about 30 external users; assistance-claims processing time cut by 70 % with renewed, AI-assisted processes.
| Organisation | LifeOne (UnoGroup) |
|---|---|
| Sector | Assistance services, Central America |
| Period | 2025–2026 · remote |
| Role | Analytics Lead — hands-on data engineering. Built and lead a three-person data team (hired and manage two data engineers), reporting to the CTO; about 60 % of my time is spent building. |
| Stack | Python · Prefect · DuckDB · Parquet · AWS (ECS, Lambda, EC2, PostgreSQL) · Docker · Power BI |
Platform
- Built end to end on AWS — an ECS cluster for the Prefect tasks and ETL images, Lambda, EC2 and PostgreSQL databases — with DuckDB over Parquet as the analytical store, Python, Prefect and Docker.
- Four environments (dev, staging, pre-prod, prod), infrastructure as code, code review and automated deployments.
- Prefect flows integrate SQL databases, REST APIs and JSON/XML feeds, with a manual sign-off step before publication: a deliberate choice for a three-person team. The next step is data tests that make the sign-off automatic.
- Scale, said plainly: the largest table has about 4 million rows and the whole warehouse fits in under 50 MB of Parquet. Small data, engineered properly.
BI and automation
- Embedded Power BI for 45 internal and about 30 external users (company stores and partners).
- Renewed the assistance-claims process with AI-based tools, cutting claim processing time by 70 %.
Confidentiality
Architecture and relative figures only. Partner names, report artefacts and absolute KPIs are not shown.