ML Data PlatformFeaturedAirtel X Labs

Spam Detection Data Platform

The data platform behind India's first AI-powered spam detection solution, protecting 300M+ Airtel users from spam and fraud calls and SMS.

Airtel users protected: 300M+
AI-powered spam detection in India: 1st

Context

Airtel launched India's first AI-powered spam detection solution to protect its 300M+ users from spam and fraud calls and SMS. The models behind it depend on large volumes of telecom network data being integrated, processed and served reliably.

Problem

Telecom-scale data had to be turned into feature-ready datasets for ML, served from a low-latency store, and kept in sync with the analytical layer, all on shared YARN clusters.

My role

I contributed to the platform's system architecture, telecom-network data integration, feature-ready datasets for ML and Aerospike-backed serving.

Approach & architecture

Integrates telecom network data into the platform's processing layer.
Builds feature-ready datasets for the spam and fraud ML models.
Serves data from Aerospike.
Bulk dumps from Aerospike are loaded into Hive, orchestrated with Airflow on large YARN clusters.

Telecom network data flows into a processing layer built on PySpark and Ab Initio, producing feature-ready datasets in Hive for the ML models. Serving is backed by Aerospike, and bulk dumps from Aerospike are loaded back into Hive by Airflow on YARN.

Results

Contributed to India's first AI-powered spam detection solution, protecting 300M+ Airtel users.
Aerospike bulk dumps loaded into Hive on large YARN clusters, orchestrated with Airflow.

Tech stack

PySpark
Hive
Aerospike
Airflow
YARN
Ab Initio
ML features

What I'd do next

Extend data-quality and drift checks to every serving-path dataset.
Add end-to-end data lineage from network sources to model features.