Event-Driven RAG Platform — Technical Interview Task (60 min)

Propose improvements to an event-driven RAG stack on AWS (S3 for docs, AWS Batch/Fargate for ETL + embeddings, API Gateway + ECS ALB API, ECS-hosted vector DB on EFS, and Redis cache).

---

Ground Rules

• Declare assumptions (numbers) at the top.
• Be concrete about the metric(s) you’re trying to move (e.g., p95 latency, ingest throughput).
• Timebox: 60 minutes prep.

Deliverables

Provide either:

• Architecture diagram (image or ASCII) with concise annotations describing what changes and why, or
• Short explanation plus Terraform-style diffs showing how you’d implement the changes (pseudo-diffs are fine).

---

Task: Scaling under load

Scenario: Scale to 20 docs/s ingest (avg 2 MB) and 250 QPS search with p95 ≤ 200 ms.

Include:

• The most impactful bottlenecks you see, with brief metric-based justification.
• Proposed changes and expected impact on latency/throughput (order-of-magnitude estimates are enough).
• Any backpressure or degradation approach you’d use during spikes.