# M3 — Full Experience Center **Priority:** P2 → P3 → P4 **Target date:** End of April 2026 **Duration:** ~20 days post-M2 **Source:** [MOM 20260326](../phase-2/conversations/20260326-mom.md#act-iii--m1--m2--m3-milestones-locked-visualization-framework-priorities-0032--0057) --- ## Objective Deliver the **fully integrated experience center** as originally envisioned: a system that blends reactive (user-triggered) and proactive (autonomous) modes, orchestrates multiple data sources, maintains real-time data refresh, and operates end-to-end in production. This is the culmination milestone — the complete Enterprise Brain decision-intelligence platform for CXOs. --- ## What Ships ### Functional Deliverables - **Fully integrated experience center:** Reactive + proactive modes working seamlessly - **Multi-data source orchestration:** Queries synthesize insights across Invictus, Gmail, Jira, and future sources - **Real-time data refresh pipeline:** Daily org data storage with RBAC - **Proactive insights:** System autonomously surfaces alerts and recommendations without user query - **Production-ready architecture:** All technical debt from M1/M2 resolved ### Core System Components - **RT core agent (full capabilities):** - Cross-source snapshot synthesis - RBAC for daily org data storage - Integration with two red boxes + one pink box - User identity and preference management - Question mode routing (ad hoc, training, data refresh) - **Experience center engine:** - Autonomous insight detection - Proactive alert generation - Contextual dashboard composition - Time-based and event-based triggers - **Multi-source data orchestration:** - Unified query layer across data sources - Cross-source relationship mapping - Conflict resolution for overlapping data - Source prioritization logic --- ## Priority Breakdown (Explicitly Decided) Naveen was explicit about the priority order within M3: ### P2: Full Experience Center Integration **Focus:** Blending reactive and proactive modes **Deliverables:** - Proactive insight engine operational - User receives both: - Reactive answers (when they ask) - Proactive alerts (when system detects) - Dashboard dynamically composed based on proactive triggers - Experience center feels autonomous, not just responsive **Success metric:** System surfaces a critical insight (e.g., production bottleneck) before user asks for it. --- ### P3: Multi-Data Source Orchestration **Focus:** Querying and synthesizing across multiple sources **Deliverables:** - Unified query layer across Invictus, Gmail, Jira - Cross-source relationship mapping (e.g., Jira ticket → email thread → production data) - Single query can trigger multi-source synthesis - Conflict resolution when sources disagree **Example query:** _"What delayed the Q1 deliverable for Project Phoenix?"_ **Expected orchestration:** 1. Jira agent: Identify delayed tasks and assignees 2. Gmail agent: Find email threads discussing delays 3. Invictus agent: Check if production dependencies were late 4. RT core: Synthesize timeline showing root cause across sources **Success metric:** Multi-source query returns coherent synthesized answer with traceability to each source. --- ### P4: Data Refresh Pipeline **Focus:** Real-time data ingestion and RBAC **Deliverables:** - Daily org data snapshot refresh - RBAC controls for data access by user role - Data staleness indicators - Refresh scheduling and monitoring - Historical snapshot storage for trend analysis **Technical requirements:** - Orange box (PostgreSQL) fully operational - Architectural databases set up - Data source APIs stable and monitored - RBAC policies defined and enforced **Success metric:** Data visible in UI is never more than 24 hours old; users see only data they have permission to access. --- ## Architecture Scope (Complete RT Core Agent) The RT core agent assumes full responsibilities deferred from M1/M2: ### Cross-Source Snapshot Synthesis - Aggregate data from multiple sources into unified snapshot - Maintain temporal consistency (align timestamps) - Resolve entity references across sources (e.g., user "nvputtagunta" in Jira = "naveen.puttagunta@..." in Gmail) ### RBAC for Daily Org Data Storage - User roles mapped to data access permissions - Row-level security for sensitive data - Audit trail for data access - Dynamic filtering based on logged-in user ### Integration with Red Boxes + Pink Box - **Two red boxes:** [Document specific components when architecture finalized] - **One pink box:** [Document specific component when architecture finalized] - Communication protocols established - Fallback behavior if integration fails --- ## Cross-Cutting Architectural Cleanup M3 resolves all technical debt accumulated in M1/M2: ### Orange Box (PostgreSQL for Instructions) - **M1 debt:** System prompts hardcoded in Python - **M3 resolution:** All agent instructions, training updates, system prompts stored in PostgreSQL - **Benefit:** Runtime updates without code deployment; centralized instruction management ### Orchestration Wiring - **M1 debt:** Orchestration being "stitched together" - **M3 resolution:** Clean, testable orchestration layer with dependency injection - **Benefit:** Reliable, maintainable, production-grade routing ### Architectural Databases - **M1 debt:** Databases not fully set up - **M3 resolution:** All databases operational: prompts, user data, snapshots, RBAC, audit logs - **Benefit:** Scalable, secure, performant data layer --- ## Parallel Workstreams ### Workstream 1: Experience Center Engine **Owner:** Yeshwanth + product team **Tasks:** - [ ] Define proactive trigger logic (time-based, event-based, threshold-based) - [ ] Implement autonomous insight detection algorithms - [ ] Build alert generation and delivery system - [ ] Create contextual dashboard composition logic - [ ] Test proactive mode with real user scenarios --- ### Workstream 2: Multi-Source Orchestration **Owner:** Backend team + data engineers **Tasks:** - [ ] Build unified query layer API - [ ] Implement cross-source entity resolution - [ ] Define conflict resolution policies - [ ] Create source prioritization logic - [ ] Test multi-source queries end-to-end --- ### Workstream 3: Data Refresh Pipeline **Owner:** Data engineers + backend team **Tasks:** - [ ] Implement daily snapshot refresh jobs - [ ] Build RBAC enforcement layer - [ ] Create data staleness monitoring - [ ] Implement historical snapshot storage - [ ] Test refresh under load --- ### Workstream 4: Technical Debt Resolution **Owner:** Backend team **Tasks:** - [ ] Migrate all hardcoded prompts to orange box - [ ] Refactor orchestration wiring for production - [ ] Set up all architectural databases - [ ] Integrate red boxes + pink box - [ ] Comprehensive integration testing --- ## Definition of Done ✅ **Functional:** - [ ] System autonomously surfaces insights without user query (proactive mode) - [ ] User can ask questions and receive answers (reactive mode) - [ ] Single query can synthesize data from multiple sources - [ ] Dashboard dynamically composed based on context and triggers - [ ] Data refresh pipeline updates data daily - [ ] RBAC restricts data access based on user role ✅ **Technical:** - [ ] RT core agent manages cross-source synthesis - [ ] Orange box stores all instructions and training - [ ] All architectural databases operational - [ ] Two red boxes + one pink box integrated - [ ] RBAC policies enforced at query time - [ ] Data staleness never exceeds 24 hours ✅ **Quality:** - [ ] System passes end-to-end integration tests - [ ] Load testing confirms performance under concurrent users - [ ] Security audit validates RBAC implementation - [ ] All M1/M2 technical debt resolved - [ ] Production deployment runbook complete ✅ **Documentation:** - [ ] System architecture fully documented - [ ] API contracts for all components finalized - [ ] User guide for reactive + proactive modes - [ ] Admin guide for RBAC configuration - [ ] Runbook for data refresh monitoring --- ## Success Criteria ### Demo Scenario 1: Proactive Insight **Setup:** User logs in Monday morning **System behavior (autonomous):** 1. RT core detects production bottleneck in Invictus data (automated analysis) 2. Cross-references with Jira: related task is delayed 3. Checks Gmail: no communication about delay in last 48 hours 4. **Proactive alert:** _"Production bottleneck detected in Line 3. Related task STEEL-421 delayed by 2 days with no recent communication. Recommend immediate follow-up."_ 5. Dashboard dynamically composes: - Production trend chart showing bottleneck - Jira task card with delay timeline - Recommended action: contact task owner **User action:** None required; insight surfaced proactively --- ### Demo Scenario 2: Multi-Source Reactive Query **User asks:** _"Why was the Q1 Phoenix deliverable delayed?"_ **System orchestration:** 1. RT core routes to Jira agent: Find task "Phoenix Q1 deliverable" and dependencies 2. Jira agent responds: Task PHOENIX-101 delayed due to dependency STEEL-421 3. RT core routes to Gmail agent: Find emails about STEEL-421 4. Gmail agent responds: Email thread shows supplier issue caused delay 5. RT core routes to Invictus agent: Check production logs for supplier issue 6. Invictus agent responds: Raw material shortage confirmed in production logs 7. **RT core synthesizes:** _"Phoenix Q1 deliverable delayed due to raw material shortage from Supplier X. Production logs show shortage started March 5. Email thread confirms supplier communicated delay March 7. Jira task STEEL-421 blocked as result."_ **UI renders:** - Timeline visualization showing cascade - Links to Jira task, email thread, production logs - Supplier status dashboard widget --- ### Demo Scenario 3: RBAC Data Filtering **Setup:** Two users with different roles **User 1 (CXO role):** Can see all production data, all emails, all Jira tasks **User 2 (Plant manager role):** Can see only Plant 3 production data, own emails, assigned Jira tasks **Query (same for both):** _"Show me production status"_ **System behavior:** - RT core checks user role - Applies RBAC filters before querying data sources - User 1 sees: All plants, all tasks, all insights - User 2 sees: Only Plant 3, only own tasks, filtered insights **Success metric:** User 2 cannot access User 1's data even if they craft custom queries. --- ## Quality Bar **Performance:** - Reactive query response time < 10 seconds (95th percentile) - Proactive insight detection latency < 5 minutes from trigger event - Data refresh completes within 4-hour window - System supports 50 concurrent users without degradation **Reliability:** - 99.5% uptime during business hours - Graceful degradation if one data source unavailable - Automatic retry with exponential backoff for failed queries - All errors logged and alertable **Security:** - RBAC enforcement validated by external audit - No unauthorized data access in penetration testing - All API endpoints authenticated and authorized - Sensitive data encrypted at rest and in transit --- ## Risks & Mitigations | Risk | Likelihood | Impact | Mitigation | | -------------------------------------------- | ---------- | -------- | ---------------------------------------------------------------------- | | Multi-source orchestration logic too complex | High | Critical | Phased rollout: 2 sources first, then 3+; simplify conflict resolution | | RBAC policy definition incomplete | Medium | High | Engage security team early; use existing org RBAC as baseline | | Data refresh pipeline performance issues | Medium | High | Incremental refresh, not full snapshot; optimize queries | | Red/pink box integration breaks | Medium | Critical | Stub interfaces early; integration testing throughout sprint | | Proactive insight logic generates noise | High | Medium | Conservative thresholds initially; user feedback loop for tuning | | Timeline too aggressive | High | Critical | Flex end-of-April to early May if needed; communicate early | --- ## Post-M3 Roadmap Once M3 ships, the platform is production-ready. Next priorities: 1. **Onboarding additional data sources:** HR systems, financial data, external market feeds 2. **Advanced trainer features (arbback):** Multi-user training, conflict resolution, training analytics 3. **Mobile experience:** Responsive design, native apps, push notifications for proactive alerts 4. **Analytics layer:** Usage patterns, query performance, insight effectiveness metrics 5. **Customization:** User-defined dashboards, custom alert rules, personalized insights --- ## Related Work **Todos from MOM:** - [ ] Connect with Watana for LLM compatibility solution — Yeshwanth, ongoing - [ ] Work with Gopal to drive operational rollout and shift estimation to story points — Yeshwanth, ongoing **Related milestones:** - [M1 — Reactive Copilot](M1-reactive-copilot.md) — Foundation - [M2 — Dynamic Workspace](M2-dynamic-workspace.md) — AI-driven UI + agent training **Related docs:** - [User Interaction Modes](user-interaction-modes.md) — Reactive vs proactive framing - [Data Flow Understanding](../phase-2/data-flow-understaning.md) — Architecture context - [Alert Lifecycle Backend](../phase-2/alert-lifecycle-be.md) — Proactive alert implementation --- **Last updated:** March 27, 2026