Planning Requirements - huqianghui/AI-Coach-vibe-coding GitHub Wiki

Requirements Traceability

Auto-generated from .planning/REQUIREMENTS.md
Last synced: 2026-04-02

Requirements: AI Coach Platform (BeiGene)

Defined: 2026-03-24 Core Value: MRs can practice realistic conversations with AI-powered digital HCPs and receive immediate, multi-dimensional feedback to improve their communication skills — anytime, without needing a real HCP or trainer.

Guiding Principles

Architecture-first: configurable, pluggable, component-based is the highest priority. Features may be incomplete, but config options, interfaces, and extension points MUST exist from day 1. Common components are reusable — changes happen at component level, not page level.

v1 Requirements

Architecture (ARCH)

  • ARCH-01: System uses pluggable adapter pattern for all AI services (LLM, STT, TTS, Avatar) — providers can be swapped via configuration without code changes
  • ARCH-02: All features are component-based and configurable — feature toggles control availability per deployment
  • ARCH-03: Backend services use dependency injection — any service can be replaced with a mock, alternative provider, or upgraded implementation
  • ARCH-04: Frontend uses shared design system component library extracted from Figma Design System — all pages reuse common components with property variations
  • ARCH-05: Azure service connections are configurable per environment — endpoints, keys, models, regions are config, not code

User Interface (UI)

  • UI-01: Shared component library based on Figma "Design System for SaaS" — buttons, cards, inputs, charts, navigation as reusable components
  • UI-02: Login page and app layout shell implemented from Figma "Design Login and Layout Shell" — sidebar navigation, header, responsive shell
  • UI-03: F2F HCP Training page implemented from Figma "F2F HCP Training Page Design" — chat area, HCP display, controls, coaching hints panel
  • UI-04: MR Dashboard implemented from Figma "Medical Representative Dashboard" — score overview, recent sessions, skill radar chart
  • UI-05: Scenario Selection page implemented from Figma "Scenario Selection Page Design" — scenario cards, filters, difficulty indicators
  • UI-06: Additional pages (admin, config, reports, session history) follow same design principles as Figma pages — self-developed using shared components
  • UI-07: All UI text externalized via react-i18next — Chinese (zh-CN) and English (en-US) supported from day 1

Authentication (AUTH)

  • AUTH-01: User can log in with simple username and password
  • AUTH-02: User session persists across browser refresh via JWT
  • AUTH-03: Two roles: User (MR) and Admin — role-based route protection
  • AUTH-04: Auth module uses dependency injection — architecture ready for Azure AD, Teams SSO, or Enterprise WeChat integration later

HCP & Scenario Configuration (HCP)

  • HCP-01: Admin can create and edit HCP profiles with name, specialty, personality type, emotional state, communication style, knowledge background
  • HCP-02: Admin can define typical objections and interaction rules per HCP profile
  • HCP-03: Admin can create and edit training scenarios with product, therapeutic area, key messages, and difficulty level
  • HCP-04: Admin can assign HCP profiles to scenarios and configure scoring dimension weights per scenario
  • HCP-05: Admin can set pass/fail threshold per scenario with weighted scoring criteria totaling 100%

F2F Coaching Simulation (COACH)

  • COACH-01: User can start a text-based F2F coaching session with an AI-powered HCP based on selected scenario
  • COACH-02: AI HCP responds in character (personality, knowledge, objections) as defined by HCP profile and scenario context
  • COACH-03: System tracks key message delivery in real-time — checklist shows which messages were delivered and which were missed
  • COACH-04: User can use voice input (Azure Speech STT) — speech recognized and sent as text to AI HCP (zh-CN + en-US)
  • COACH-05: AI HCP responses are spoken via Azure Speech TTS — natural-sounding voices in Chinese and English
  • COACH-06: Voice interaction supports GPT Realtime API (WebSocket) for sub-1s conversational latency as configurable premium option
  • COACH-07: Azure AI Avatar renders digital human visual for HCP as configurable premium option — falls back to TTS-only when disabled or unavailable
  • COACH-08: Real-time coaching hints displayed in side panel during conversation — contextual suggestions based on conversation progress
  • COACH-09: Conversations are immutable once completed — only scoring and feedback can be added after completion

Scoring & Feedback (SCORE)

  • SCORE-01: System scores completed sessions across 5-6 configurable dimensions (key message delivery, objection handling, communication skills, product knowledge, scientific accuracy)
  • SCORE-02: Scoring uses Azure OpenAI to analyze conversation transcript against scenario criteria and HCP expectations
  • SCORE-03: Post-session feedback report shows strengths and weaknesses per dimension with specific conversation quotes
  • SCORE-04: Post-session feedback includes actionable improvement suggestions per dimension
  • SCORE-05: Scoring dimension weights are configurable per scenario — admin sets relative importance via weighted sliders

Conference Mode (CONF)

  • CONF-01: User can start a conference presentation mode with multiple virtual HCP audience members
  • CONF-02: Multiple AI HCPs ask questions from audience — questions queue with turn management
  • CONF-03: Conference session includes live transcription display
  • CONF-04: Conference sessions are scored using the same multi-dimensional scoring system as F2F

Content Management (CONTENT)

  • CONTENT-01: Admin can upload training materials (PDF, Word, Excel) organized by product and therapeutic area
  • CONTENT-02: Uploaded materials feed into AI knowledge base for more accurate HCP simulation (RAG-style grounding)
  • CONTENT-03: Training materials support versioning and folder organization

Analytics & Reports (ANLYT)

  • ANLYT-01: User can view session history — list of past sessions with date, scenario, score, duration
  • ANLYT-02: User can view personal performance trends — score improvement over time per dimension
  • ANLYT-03: Admin can view organization-level analytics — BU comparisons, skill gap heatmaps, training completion rates
  • ANLYT-04: System recommends next training scenarios based on user's scoring history and identified weaknesses
  • ANLYT-05: Reports and dashboards use Recharts radar/spider charts for multi-dimensional score visualization

Platform (PLAT)

  • PLAT-01: i18n framework (react-i18next) integrated from day 1 — all UI strings externalized, zh-CN and en-US
  • PLAT-02: Responsive web design — same app works on desktop, tablet, mobile, and Teams Tab (iframe)
  • PLAT-03: Admin can configure Azure service connections (OpenAI, Speech, Avatar, Content Understanding) from web UI with connection testing
  • PLAT-04: Per-region deployment supported — single codebase, per-region configuration for data residency (China, EU)
  • PLAT-05: Voice interaction mode (STT/TTS vs GPT Realtime vs Voice Live) configurable per deployment and per session

v2 Requirements

Extended Authentication

  • AUTH-V2-01: Azure AD (Entra ID) SSO integration
  • AUTH-V2-02: Microsoft Teams SSO for Teams Tab embedding
  • AUTH-V2-03: Enterprise WeChat (企业微信) SSO integration
  • AUTH-V2-04: Complex role hierarchy (DM, MSL, BU Head, Regional Director)

Extended Features

  • EXT-01: Teams Bot integration (full Bot Framework, adaptive cards)
  • EXT-02: WeChat Mini Program frontend
  • EXT-03: PDF/Excel export for reports and session transcripts
  • EXT-04: Azure Voice Live API as unified premium voice+avatar path
  • EXT-05: Custom analyzer configuration for Content Understanding

Out of Scope

Feature Reason
Live video conferencing Training simulator, not video call — use Avatar for visual
Mobile native app (iOS/Android) Responsive web covers mobile; company phones have browsers
Multi-tenancy Single tenant per-region; deploy separate instances if needed
Gamification (leaderboards, badges) Professional training, not a game; focus on improvement trends
Conversation editing/retry from mid-point Breaks simulation realism; MRs can "Try Again" with new session
Custom LLM fine-tuning Prompt engineering + RAG achieves 90% of value at lower cost
Real-time collaborative training Edge case; DMs review scores after, not observe live

Traceability

Requirement Phase Status
ARCH-01 Phase 1 Complete
ARCH-02 Phase 1 Complete
ARCH-03 Phase 1 Complete
ARCH-04 Phase 1 Complete
ARCH-05 Phase 1 Complete
UI-01 Phase 1 Complete
UI-02 Phase 1 Complete
UI-03 Phase 2 Complete
UI-04 Phase 4 Complete
UI-05 Phase 2 Complete
UI-06 Phase 4 Complete
UI-07 Phase 1 Complete
AUTH-01 Phase 1 Complete
AUTH-02 Phase 1 Complete
AUTH-03 Phase 1 Complete
AUTH-04 Phase 1 Complete
HCP-01 Phase 2 Complete
HCP-02 Phase 2 Complete
HCP-03 Phase 2 Complete
HCP-04 Phase 2 Complete
HCP-05 Phase 2 Complete
COACH-01 Phase 2 Complete
COACH-02 Phase 2 Complete
COACH-03 Phase 2 Complete
COACH-04 Phase 3 Complete
COACH-05 Phase 3 Complete
COACH-06 Phase 3 Complete
COACH-07 Phase 3 Complete
COACH-08 Phase 2 Complete
COACH-09 Phase 2 Complete
SCORE-01 Phase 2 Complete
SCORE-02 Phase 2 Complete
SCORE-03 Phase 2 Complete
SCORE-04 Phase 2 Complete
SCORE-05 Phase 2 Complete
CONF-01 Phase 3 Complete
CONF-02 Phase 3 Complete
CONF-03 Phase 3 Complete
CONF-04 Phase 3 Complete
CONTENT-01 Phase 3 Complete
CONTENT-02 Phase 3 Complete
CONTENT-03 Phase 3 Complete
ANLYT-01 Phase 4 Complete
ANLYT-02 Phase 4 Complete
ANLYT-03 Phase 4 Complete
ANLYT-04 Phase 4 Complete
ANLYT-05 Phase 4 Complete
PLAT-01 Phase 1 Complete
PLAT-02 Phase 1 Complete
PLAT-03 Phase 2 Complete
PLAT-04 Phase 1 Complete
PLAT-05 Phase 1 Complete

Coverage:

  • v1 requirements: 52 total
  • Mapped to phases: 52
  • Unmapped: 0

Requirements defined: 2026-03-24 Last updated: 2026-03-24 after roadmap creation -- all requirements mapped to phases