Atul Rai
Back to Projects

Enterprise Data Platform

Centralized DataMart on AWS Aurora with ingestion, governance, and quality pipelines powering ML workflows across teams.

FastAPI
AWS Aurora
Airflow
DBT
EMR
PostgreSQL
Kafka
Pandas

+13%

ML Accuracy

Aurora

Core Store

Lead

Role

Architecture

Overview

At Aideo Technologies / Rule14, I led development of core data platform features over 3.5 years — designing scalable FastAPI services, architecting a centralized DataMart on AWS Aurora, and owning customer-facing ingestion, transformation, and egestion pipelines at scale.

Problem

Data science teams were blocked by fragmented data access, inconsistent quality, and no unified governance layer. External system integrations following a strategic alliance added complexity — data had to flow reliably without compromising downstream ML accuracy.

Architecture

External systems feed ingestion pipelines → centralized DataMart on AWS Aurora → governance and validation layer → ML workflows and customer-facing APIs. See the interactive diagram above.

Challenges

  • Data accessibility: Scientists needed a single, reliable source instead of ad-hoc extracts from multiple systems
  • Quality at scale: Bad data silently degraded ML model performance with no early warning
  • Integration reliability: Post-alliance ingestion pipelines had to handle diverse external formats without breaking existing workflows
  • API performance: Platform APIs needed to serve both batch analytics and real-time customer operations

Scale / Performance

  • Centralized DataMart significantly improved data accessibility and accelerated Data Scientist workflows across teams
  • Governance pipelines with anomaly detection and alerting improved downstream ML model accuracy by 13%
  • Owned end-to-end ingestion → transformation → egestion pipelines with reliable data flow at scale

Interesting Engineering Decisions

AWS Aurora as the DataMart core

Aurora provided the durability and query performance needed for a centralized analytics layer while integrating cleanly with existing AWS infrastructure (EMR, Lambda, Airflow).

Governance pipelines over manual QA

Automated validation rules and anomaly detection caught data quality issues before they reached ML training — more scalable than manual review and measurable in model accuracy gains.

FastAPI for platform APIs

FastAPI's async performance and type safety fit the platform's need for both high-throughput ingestion endpoints and customer-facing egestion APIs.

Lessons Learned

  • Data governance ROI shows up in ML metrics, not just ops dashboards — the 13% accuracy gain made the case
  • Centralizing data access accelerates every downstream team, not just data science
  • Ingestion pipeline ownership means owning failures too — alerting and anomaly detection are non-negotiable

Outcome

Built the platform's core data infrastructure: centralized DataMart, governance pipelines, and reliable cross-system ingestion — enabling faster ML iteration and measurable model accuracy improvements (Jan 2022 – Aug 2025).