Data CloudExpertagentforceretrievalpersonalization
Evaluate Agentforce context retrieval from Data Cloud profiles
Real World Scenario
Service Agentforce agent needs customer context from unified profile; architects must define what attributes are safe for LLM retrieval vs masked.
Expected Answer
• Classify profile attributes: retrievable, masked, prohibited for LLM context
• Use retrieval filters enforcing consent and role of requesting user
• Minimize context window — summary insight not raw event history dump
• Log retrieved attributes per agent session for audit
• Refresh context on profile update during long conversations
• Fallback when profile incomplete — agent asks clarifying questions
• Test prompt injection via profile field values sanitized on ingest
Follow-Up Questions & Answers
Click to expand — each follow-up includes a direct, interview-ready answer
Direct answer: Classify profile attributes: retrievable, masked, prohibited for LLM context Also consider: Use retrieval filters enforcing consent and role of requesting user In practice: Minimize context window — summary insight not raw event history dump Validate against production-scale constraints and long-term platform evolution.
Architect Perspective
Data Cloud + Agentforce boundary defines trust — not every profile field belongs in LLM context.