Alara — enterprise AI platform
Backend architecture for Alara, an enterprise AI platform used by major banks: FastAPI microservices, Kafka/Redis event pipelines and GCP deployments, plus product ownership. Client details are confidential.
- 01Bank users & systems
- 02FastAPI services
- 03Kafka / Redis events
- 04AI workloads on GCP
Problem
Banks don't buy AI demos. An AI platform for major banks has to fit regulated environments: clear service boundaries, predictable behaviour under load, auditable data flows and secure cloud deployments. The backend had to carry ML workloads without becoming the bottleneck, and someone had to turn business and ML requirements into a roadmap engineers could build.
What I built
- Backend architecture for the platform, as backend lead.
- Python (FastAPI) microservices designed for regulated environments.
- Event-driven pipelines on Kafka and Redis that cut latency and raised throughput.
- API design and data models across the services.
- Cloud deployments on GCP.
- Product ownership: translating business and ML requirements into a technical roadmap.
Stack
- Python
- FastAPI
- Kafka
- Redis
- GCP
- Microservices
- REST APIs
Result
- In use by major banks.
- Lower latency and higher throughput after moving work onto event-driven pipelines.
- One owner for both the architecture and the roadmap, so product decisions and technical constraints met early.
Engagement
Enterprise engagement
My role
Since 2024 I’ve been Senior Backend Lead and Product Owner at WAI Industries (ISSM.ai), where I lead backend architecture for Alara, an enterprise AI platform used by major banks. The specifics of clients, models and data stay confidential; this page covers how the backend is built and what I own.
Architecture
- Microservices in Python (FastAPI). Each capability is its own service with a clear API contract, so teams can change one part without destabilising the rest, and each service can be reviewed, scaled and secured on its own.
- Event-driven pipelines on Kafka and Redis. Long-running and ML-heavy work moves through events instead of blocking request/response calls. That change reduced latency for users and increased throughput for the platform.
- APIs and data models. I own the API design and the data models across services, which is where most long-term complexity in a platform like this lives.
- GCP deployments. Services are deployed and run on Google Cloud.
Built for regulated environments
Working with banks shapes every decision. Security, observability and correctness are requirements, not polish: services have narrow responsibilities, data flows are explicit, and the platform has to fail predictably rather than cleverly. My view is that AI systems need strong infrastructure before intelligence, and that is the part of Alara I’m responsible for.
Product ownership
Alongside architecture I act as Product Owner: I translate business and ML requirements into a technical roadmap, decide what gets built in what order, and make sure what the bank-side stakeholders need and what the backend can safely deliver meet early instead of at release time.