Skip to main content

Serverless Data Processing

Process streaming data events in real-time without managing servers.

Architecture Overview​

An event-driven architecture that scales to zero when idle and handles traffic spikes automatically.

Use Cases​

  • Real-time Image Classification: Uploaded images trigger a function to run inference and store tags.
  • IoT Anomaly Detection: Sensor data streams are analyzed for out-of-bounds values.
  • Chatbots: User messages trigger LLM inference functions on demand.

Components​

  1. Event Source: The trigger for the pipeline (HTTP request, file upload, or message queue).
  2. Function-as-a-Service (FaaS): Substrate's Serverless GPU containers that spin up in milliseconds to process the request.
  3. Database: A scalable NoSQL store (like Cassandra or MongoDB) to save the results.

Advantages​

  • Cost Efficiency: Pay only for the milliseconds the code is running.
  • Scalability: Automatically handles concurrent requests without manual provisioning.
  • Simplicity: Developers focus on code (handler.py), not infrastructure.