Data Eng2025
AMFI ETL Pipeline
An end-to-end ETL pipeline ingesting, transforming and loading AMFI regulatory data into an analytics warehouse, orchestrated with Apache Airflow and surfaced through Power BI dashboards. Built during my Data Engineering internship at Intellect Design Arena.
- Regulatory reporting cycle
- Automated
- Pipeline stages with quality gates
- 4
- Stakeholder-facing dashboards
- Power BI
The problem
Regulatory data from the Association of Mutual Funds in India is public, structured, and almost unusable in practice — inconsistent formatting, siloed sources, and no guarantee that today's file looks like yesterday's. Analysts were spending their time on manual preparation rather than analysis, and manual preparation is exactly where silent errors enter a dataset nobody later questions.
The approach
Automate the whole path and make the pipeline assert its own correctness. Airflow orchestrates ingest, transform and load on a schedule; automated data quality checks sit between the stages so a malformed upstream file fails loudly instead of quietly poisoning the warehouse. Power BI dashboards sit on the clean end for stakeholders.
Financial regulators produce vast amounts of structured data that remains trapped in siloed, poorly formatted sources. I wanted to build a pipeline that turns raw AMFI data into a reliable, analysis-ready dataset that financial engineers can actually trust.
Architecture
- 01
Ingest
Scheduled extraction of AMFI regulatory data from its published sources.
- 02
Quality gates
Automated checks between stages — a malformed upstream file fails the DAG rather than reaching the warehouse.
- 03
Transform & load
Normalisation into a structured analytics warehouse in PostgreSQL.
- 04
Reporting
Power BI dashboards over the clean dataset for stakeholder consumption.
Stack
- Orchestration
- Apache Airflow
- Processing
- Python · Pandas · NumPy
- Warehouse
- PostgreSQL · SQL
- Python
- Apache Airflow
- ETL
- SQL
- PostgreSQL
- Power BI
- Pandas
- NumPy
Outcome
Regulatory reporting that runs itself, with a significant reduction in manual data preparation time and quality gates that surface upstream problems instead of hiding them. Shipped as an internship deliverable and used by stakeholders.