Scaling AI with Event-Driven Microservices

When building CareerLens, we realized that synchronous API calls to LLMs (like Gemini) create massive bottlenecks.

The Solution

We adopted an event-driven architecture using Pub/Sub mechanisms.

1. Upload: User uploads a resume.
2. Event: 'resume.uploaded' event is published.
3. Workers: Multiple microservices pick up the event to extract skills, generate embeddings, and query the LLM concurrently.

This reduced processing time from 30 seconds to under 5 seconds for complex multi-agent workflows.