Skip to main content

S2. Persona Chatbot Across 5 Departments (AMWAY)

Published 2026-05-14Updated 2026-06-303 min read

1. URL & Personas

  • /chat · All personas

2. User Stories

  • P1 (ABO Field): "Top 10 ABOs who reached Diamond this quarter and their Downline activity"
  • P5 (Compliance): "Protein-campaign ad review + minor block counts"

3. UI

  • Left: chat / Right: SSE tool-call trace
  • Persona switcher + four suggested queries

4. 12 Tools

ToolDescription
semantic_searchOpenSearch BM25 + KNN + Rerank (multilingual)
neptune_queryopenCypher
abo_tree_queryABO tree depth query (S9-A)
subscription_lifecycleSubscription state · churn signature (S10-A)
campaign_simulatorBayesian ROAS
behavior_change_detectStatistical signature
social_trend_joinSocial multilingual
weather_joinKMA · global weather
economic_joinFX · prices (multi-country)
regulation_joinDirect-selling regulations
cluster_runKMeans + LLM labeling
pii_maskPII masking

5. Per-persona Variations

Psystem prompt keywordsPrimary tools
P1ABO · Sponsor · Downline · PV/BVabo_tree_query, campaign_simulator
P2Cohort · LTV · SOVcluster_run, social_trend_join
P3Accuracy · driftbehavior_change_detect, neptune_query
P4Level · NPS · subscriptionsubscription_lifecycle, semantic_search
P5Regulation · consent · minorsregulation_join, pii_mask

6. Guardrails

  • Input: 4 topics + direct selling + minors + HFF
  • Output: PII masking; citations are mandatory for regulatory answers
  • Tools: write tools are restricted to P3 and P5

7. Demo Scenarios

  1. (P1) "Top 10 ABOs reaching Diamond this quarter" → abo_tree_query + neptune_query
  2. (P1 → P4 switcher) Same query → tone shifts (level / NPS lens)
  3. (P5) "Review Nutrilite advertising" → regulation_join + list of high-risk SKUs