Consaltica. Building a Modern Anti-Fraud Platform
On this project I served as Chief Solution Architect and product lead, responsible for defining the platform's target architecture, its technology strategy, and the solution's alignment with business objectives.
My remit covered architectural design, technology stack selection, building end-to-end data flows, and balancing speed of delivery, scalability, and operational risk.
Platform Vision and Purpose
The platform is designed as a universal foundation for building high-load decision-making systems operating in real time and in batch mode. The main focus was on creating a product that serves the needs of the business, analysts, and engineering teams equally well.
The solution is aimed at anti-fraud, scoring, compliance, processing, and other data-driven domains where controllable logic, low latency, and the ability to evolve without architectural compromises are critical.
Data Architecture and ETL
The data architecture is built around an event-driven approach, with Apache Kafka serving as the central integration and event layer. In the product vision, Kafka is used not just as a transport but as a foundation for scaling, domain isolation, and reducing coupling between components.
Apache Spark handles complex ETL pipelines, transformations, aggregations, and analytical calculations, providing a unified approach to streaming and batch data processing.
Real-Time Decision-Making Engine
The real-time decision engine is implemented in Java, with Hazelcast as a distributed in-memory layer. This architectural choice made it possible to:
- deliver minimal latency under high transaction loads;
- achieve linear horizontal scalability;
- isolate critical business logic from the analytics and ETL pipelines.
As Chief Solution Architect and Product Owner, I made the key decisions on service boundaries and SLAs for various decision-making scenarios.
Orchestration and Logic Management
Camunda serves as the orchestration and business process management layer. This made it possible to move a significant share of the decision-making logic out of code and into a managed process model, reducing time-to-market and operational risks.
From an architectural and product standpoint, this enabled the platform to scale not only technically but also organizationally, by dividing responsibilities between teams.
Analytics, Control, and Feedback
The platform includes a product analytics layer: a builder for dashboards, metrics, and aggregates, geared toward business oversight of decision quality and model performance.
PostgreSQL serves as a reliable store for data and metadata, supporting auditing, versioning, and transparency of changes, which are critical requirements for enterprise solutions.
Scaling and Product Evolution
Architecturally, the platform was designed from the outset as an extensible product:
- support for new channels and data sources without changing the core;
- readiness for ML models and advanced analytics;
- the ability to adapt to different business domains without accumulating architectural debt.
Architectural decisions were made with long-term product development, total cost of ownership, and the maturity of operational processes in mind.
Technology Stack
- Event & Data Streaming: Apache Kafka
- ETL & Analytics: Apache Spark
- Decision Engine: Java, Hazelcast
- Process Orchestration: Camunda
- Data & Metadata Storage: PostgreSQL
- Architecture: event-driven, distributed systems, real-time & batch processing