Case study 03 · Comcast
Real-Time Intelligence Architecture
Event-driven, microservices-based architecture using Kafka, Databricks and Apache Spark for real-time analytics, fraud and anomaly detection, recommendations, risk scoring and workflow automation.
- Company
- Comcast
- Role
- AI Solutions Architect
- Timeframe
- Oct 2021 – Apr 2024
- Location
- United States
- Kafka
- Databricks
- Spark
- Microservices
- Real-Time AI
01The challenge
High-volume operational ecosystems generate events across customer systems, CRM, order management and operational platforms. Intelligence such as fraud or anomaly detection loses value when it arrives late or sees only part of the picture.
The architecture had to move data reliably between systems, make it available for analytics and models, and feed results back into workflows, all at scale and with operational resilience.
02Keerthi's role
Keerthi designed the integration and intelligence architecture as AI Solutions Architect, from event streaming and data ingestion through microservices and deployment.
- Designed enterprise integration workflows using Kafka event streaming, Databricks, Apache Spark, APIs and cloud-native platforms.
- Developed microservices architectures for real-time analytics, fraud detection, anomaly detection, recommendation engines, risk scoring and workflow automation.
- Designed data ingestion and API integration frameworks connecting customer, operational, enterprise and analytics systems.
- Supported Kubernetes and Docker-based deployments for scalable, resilient AI workloads.
03Architecture / approach
Events are the backbone: systems publish what happens, and intelligence services subscribe to what they need.
- 01
Event streaming
Kafka carries events from CRM, order management and operational platforms, decoupling producers from the services that react to them.
- 02
Data processing
Databricks and Apache Spark process high-volume data for real-time analytics and model-driven intelligence.
- 03
Intelligence services
Microservices deliver fraud detection, anomaly detection, recommendation engines and operational risk scoring.
- 04
Closed-loop workflows
Results feed workflow automation, operational dashboards and enterprise applications through APIs.
- 05
Resilient operations
Kubernetes, Docker, CI/CD and observability keep AI workloads scalable, available and reliable.
04Technology
Streaming & Data
- Kafka
- Databricks
- Apache Spark
- SQL
- Python
Architecture
- Microservices
- Event-Driven Architecture
- Distributed Systems
- REST APIs
Platform
- Kubernetes
- Docker
- CI/CD
Analytics
- Power BI
- Tableau
05Enterprise value
An event-driven foundation for operational intelligence across high-volume enterprise systems.
- Real-time analytics, fraud detection, anomaly detection, recommendations and risk scoring delivered as services.
- Interoperability between customer systems, operational systems, enterprise applications and analytics platforms.
- Scalable, resilient deployment of enterprise AI workloads on Kubernetes and Docker.
Outcomes are described qualitatively. Figures appear only where they are verified.
[ADD PROJECT METRIC IF AVAILABLE]06Architecture diagram
Sources
- CRM
- Order management
- Operational platforms
- Customer systems
Streaming
- Kafka event streams
- Data ingestion
- API integration
Processing
- Databricks
- Apache Spark
- SQL
Intelligence
- Fraud detection
- Anomaly detection
- Recommendations
- Risk scoring
Action
- Workflow automation
- Dashboards
- Enterprise apps
07Key takeaways
- 01
Decouple producers from consumers. New intelligence services should not require changes to source systems.
- 02
Real-time only matters if the result reaches a workflow. Design the action path, not just the model.
- 03
Operational reliability is a feature of AI systems, not a separate concern.