Perfect — this is the most important round. This is where they decide: “Can this person design Salesforce-scale systems?”

I’ll simulate a real system design interview and give you a top-tier answer you can adapt.

⸻

🎯 PROBLEM STATEMENT (Salesforce Style)

👨‍💼 Praveen:

“Design a global HR data platform for Salesforce that:

• Integrates data from multiple systems (HRIS, payroll, performance, etc.)
• Supports real-time analytics and reporting
• Enables AI-driven insights (Agentforce / copilots)
• Ensures strong governance (PII, compliance)
• Scales globally (millions of users)

Walk me through your design.”

⸻

🚀 YOUR ANSWER (STRUCTURED — USE THIS)

⸻

🧩 1. Clarify Requirements (VERY IMPORTANT)

“I’ll break this into functional and non-functional requirements.”

Functional:

• Multi-source ingestion (HRIS, payroll, etc.)
• Real-time + batch analytics
• AI-driven insights (copilot / agents)
• Self-service querying

Non-functional:

• Security (PII, GDPR)
• Scalability (global users)
• Low latency (near real-time)
• High availability

⸻

🏗️ 2. HIGH-LEVEL ARCHITECTURE

⸻

Layer 1: Ingestion Layer

• APIs (HR systems like Workday)
• CDC streams (employee updates)
• Kafka for real-time ingestion

⸻

Layer 2: Processing Layer

• Streaming → Kafka consumers
• Batch → dbt / orchestration

⸻

Layer 3: Storage Layer

• Iceberg on S3 (curated + historical)
• Source systems (for real-time federation)

⸻

Layer 4: Query Layer

• Trino (federated + analytical queries)

Layer 5: Metadata Layer

• OpenMetadata
• Tracks:
• ownership
• lineage
• sensitivity

Layer 6: Governance Layer

• OPA (policy engine)
• Row/column-level security
• Audit logs

⸻

Layer 7: AI / Agent Layer

• RAG-based system
• Agents:
• Query data
• Generate insights
• Trigger workflows

⸻

Layer 8: Consumption Layer

• Dashboards (Tableau)
• APIs
• AI copilots (Agentforce)

⸻

🧠 3. DATA FLOW (EXPLAIN CLEARLY)

1. Data enters via API / CDC
2. Stored in Iceberg OR accessed via federation
3. Metadata captured in OpenMetadata
4. Policies enforced via OPA
5. Users / AI query via Trino
6. Results returned securely

⸻

🔥 4. KEY DESIGN DECISIONS (THIS IS WHERE YOU WIN)

⸻

✅ Decision 1: Federation + Storage (Hybrid)

“Not all data is moved”

• Real-time → federation
• Historical → Iceberg

👉 Trade-off:

• Speed vs consistency

⸻

✅ Decision 2: Metadata-Driven Governance

• No hardcoded rules
• Policies driven by metadata

⸻

✅ Decision 3: AI-Ready Design

• Metadata → embeddings
• Enables RAG + agents

⸻

🔐 5. SECURITY (VERY IMPORTANT)

• PII masking
• Row-level filtering
• Access via identity (SSO, RBAC, ABAC)
• Audit trail

👉 “Security enforced at query layer”

⸻

⚡ 6. PERFORMANCE

• Partition pruning (Iceberg)
• Predicate pushdown (Trino)
• Caching / materialized views
• Resource groups

⸻

🌍 7. GLOBAL SCALE

• Multi-region deployment
• Data locality
• Regional compliance

⸻

🤖 8. AI / AGENT USE CASE

Example:

User asks: 👉 “Show me attrition risk in EU”

Flow:

• Agent retrieves metadata
• Generates SQL
• Query executed via Trino
• Results returned with governance

⸻

🔥 9. TRADE-OFFS (CRITICAL)

⸻

Decision Trade-off Federation Performance vs freshness Iceberg Storage vs latency AI agents Flexibility vs control

⸻

👉 Say: “No system is perfect — we balance trade-offs based on use case.”

⸻

🚨 10. FAILURE SCENARIOS

Be ready:

⸻

❓ “What if a source system is down?”

• Fallback to cached / Iceberg data
• SLA-based routing

⸻

❓ “What if query is slow?”

• Analyze plan
• Optimize joins
• Reduce data movement

⸻

⸻

🔥 FINAL CLOSING LINE

“The goal is not just to build a data platform, but to enable a governed, intelligent data ecosystem where both humans and AI can safely access and act on data.”

⸻

🚀 WHAT MAKES THIS ANSWER TOP 1%

You showed:

• Architecture depth ✅
• Trade-offs ✅
• AI thinking ✅
• Governance ✅
• Real-world experience ✅

⸻

🔥 NEXT STEP (HIGHLY RECOMMENDED)

Let’s do:

👉 Whiteboard pressure round

I’ll interrupt and ask:

• “Why not Snowflake?”
• “How do you handle latency?”
• “What breaks first?”

Just say: 👉 “pressure round” and we go hardcore 🚀