← Salesforce Interview Insights

Salesforce·Software Engineer·Onsite - System Design / Architecture·Senior

Senior
Jun 2026

Summary

Salesforce system design round focused entirely on building an analytics dashboard for a large LLM-based product. Heavy on metrics selection, data pipeline architecture, and the usual freshness vs cost arguments. Solid question with a lot of surface area.

Questions Asked (1)

Q1

Design an analytics dashboard for a large LLM-powered product. Walk through which metrics to surface, how to ingest and process events at scale, and how to back the dashboard with appropriate storage and query layers.

System DesignProduct Analytics & MetricsTechnical Trade-offs
Author's notes

This one sprawled in every direction fast.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying the product's goals and key user personas to prioritize metrics, then design a scalable event ingestion pipeline (e.g., Kafka) and a multi-layered storage architecture (real-time and batch). Finally, discuss query engines and trade-offs between latency, cost, and flexibility.

Pro tip: Emphasize that the dashboard should drive actionable insights, not just display data—tie each metric to a decision or alert. Also, mention the importance of data governance and privacy when handling LLM interactions.

1. Clarify Requirements and Metrics

Ask about the product's objectives, user roles, and key performance indicators (KPIs). Identify metrics like token usage, latency, error rates, user engagement, and cost per query.

2. Design Event Ingestion Pipeline

Propose a scalable ingestion layer using a distributed message queue (e.g., Kafka) to collect events from various sources. Ensure data validation, enrichment, and schema management.

3. Choose Storage and Processing Layers

Select a combination of real-time storage (e.g., Druid, ClickHouse) for low-latency queries and a data lake (e.g., S3) with batch processing (e.g., Spark) for historical analysis. Consider a time-series database for metrics.

4. Implement Query and Visualization Layer

Use a query engine (e.g., Presto, Apache Superset) that can federate across storage layers. Design the dashboard with drill-down capabilities and customizable views for different stakeholders.

5. Address Trade-offs and Scalability

Discuss trade-offs between latency, cost, and complexity. Highlight strategies for scaling (e.g., partitioning, caching) and ensuring data consistency and reliability.

Key Points to Mention

  • Metric selection: token usage, latency, error rates, user engagement, cost per query, model drift.
  • Event ingestion: Kafka for high throughput, schema registry, exactly-once semantics.
  • Storage: real-time OLAP (ClickHouse/Druid) for dashboards, data lake (S3) for raw events, time-series DB for metrics.
  • Query layer: Presto/Trino for federated queries, caching for performance.
  • Scalability: partitioning, horizontal scaling, backpressure handling.
  • Data governance: PII handling, access control, compliance (GDPR, CCPA).

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.