Persona-to-grounding mapping

The six AVX personas vary in their grounding needs. The following table summarizes which grounding surface each persona primarily uses and what data it expects at session initialization. The table is the prescriptive implementation guide; the authoritative code lives in agents/supervisor/persona_definitions.py and lambdas/api-assistant-chat/persona_map.py.

Persona	Primary segment	LTM products loaded at session init	KB grounding in use
Fleet Driver	Fleet operator	vehicle_identity, service_records, charging_sessions, ota_campaigns, vehicle_telemetry_aggregated, energy_usage	LTM + unfiltered KB (generic retrieve tool) — no pre-scoped source_category filter; the retrieve tool returns semantically matched passages across all content (e.g. DTC guides, recall bulletins, owner manual) relevant to the query
Consumer	Rental	vehicle_identity, charging_sessions, customer_360, customer_interactions, ota_campaigns, energy_usage	LTM + unfiltered KB (generic retrieve tool) — rental context, interaction history, and charging data are the primary sources; retrieve is available for policy and manual lookups
Retail Advisor	OEM	vehicle_identity, customer_360, customer_interactions	LTM + unfiltered KB (generic retrieve tool) — retail advisor queries center on customer history and vehicle sales context; retrieve is available for brand policy and warranty lookups
Vehicle Owner	OEM	vehicle_identity, service_records, charging_sessions, customer_360, ota_campaigns, vehicle_telemetry_aggregated	LTM + unfiltered KB (generic retrieve tool) — owner queries are anchored to the specific vehicle’s history and campaign status; retrieve is available for recall and manual lookups
Service Advisor	Fleet (mixed)	vehicle_identity, service_records, customer_360, customer_interactions	LTM + pre-scoped filtered KB — source categories pre-scoped via PERSONA_KB_FILTERS: tsb_recall, owner_manual, service_policy, charging_narrative; service advisors need reference content alongside the vehicle’s record to make and justify service decisions
Technician	Fleet (mixed)	vehicle_identity, service_records	LTM + pre-scoped filtered KB — source categories pre-scoped via PERSONA_KB_FILTERS: dtc_guide, parts_catalog, service_policy; technicians need diagnostic reference and parts data alongside the service record for the vehicle under investigation

All six AVX personas have KB grounding available, but the mechanism differs. Fleet Driver, Consumer, Retail Advisor, and Vehicle Owner use the generic unfiltered retrieve tool — the supervisor passes an optional source_category keyword argument or retrieves across all categories based on the query. Service Advisor and Technician use pre-scoped KB access defined in PERSONA_KB_FILTERS in persona_definitions.py, which restricts the Retrieve filter to the source categories appropriate for each role. When a mixed persona’s query does not require reference lookup — for example, "when was this vehicle last serviced?" — the supervisor omits the KB Retrieve call and answers from Memory alone, saving both latency and token cost.

Persona is always inferred server-side from validated Cognito claims, never accepted from client-supplied parameters. The mapping logic is described in Personas and Tenancy; the security rationale is covered in Security.