Product Development Plan
Macau University Library LLM Systems
Project: LLM-based Inquiry & Recommendation Systems
Client: Macau University Library
Total Effort: 65 man-days
Budget: HK$180,000
Executive Summary
This plan covers the development of two integrated LLM-based systems:
- System A: Tabular Data Inquiry System (circulation, privileges, loans)
- System B: Reference Material Recommendation System (theses, books, papers)
Both systems share a common infrastructure using DeepSeek-R1-Distill-Llama 8B deployed on-premise via Docker, with SYVA Platform as the CMS layer.
Product Architecture Overview
Architecture Diagram (Mermaid)
flowchart TB
subgraph FRONTEND["Frontend Layer"]
direction LR
CHAT["Chatbot UI<br/>Query Input | Response | Follow-up"]
ADMIN["Admin CMS - SYVA<br/>Table Mgmt | Doc Upload | Access Control"]
end
subgraph API["API Gateway"]
GW["REST API / WebSocket<br/>Real-time Chat"]
end
subgraph BACKEND["Backend Services"]
direction LR
ROUTER["Query Router<br/>Intent Detection"]
RAG["RAG Engine<br/>Retriever | Re-ranker"]
RESP["Response Generator<br/>Context Assembly | LLM Inference"]
end
subgraph DATA["Data Layer"]
direction LR
VECTOR[("Vector DB<br/>Embeddings")]
PG[("PostgreSQL<br/>Metadata")]
LLM["DeepSeek 8B<br/>On-premise Model"]
end
CHAT --> GW
ADMIN --> GW
GW --> ROUTER
GW --> RAG
GW --> RESP
ROUTER --> VECTOR
RAG --> VECTOR
RAG --> PG
RESP --> LLM
style FRONTEND fill:#e3f2fd,stroke:#1976d2
style API fill:#fff3e0,stroke:#f57c00
style BACKEND fill:#e8f5e9,stroke:#388e3c
style DATA fill:#fce4ec,stroke:#c2185b
PART 1: DEVELOPMENT TEAM PLAN
Sprint Breakdown (Recommended: 2-week sprints)
Sprint 1-2: Infrastructure & Foundation (10 days)
Deliverables
| Task | Owner | Days | Priority |
|---|---|---|---|
| Docker environment setup | DevOps | 2 | P0 |
| DeepSeek 8B model installation & testing | ML Engineer | 3 | P0 |
| SYVA Platform installation & config | Backend | 2 | P0 |
| Database schema design (PostgreSQL) | Backend | 1.5 | P0 |
| Vector DB setup (ChromaDB/Qdrant) | Backend | 1.5 | P1 |
Technical Specifications
Docker Compose Structure:
services:
deepseek-llm:
image: deepseek-r1-distill-llama-8b
deploy:
resources:
reservations:
devices:
- capabilities: [gpu]
ports:
- "8080:8080"
vector-db:
image: qdrant/qdrant
ports:
- "6333:6333"
volumes:
- vector_data:/qdrant/storage
postgres:
image: postgres:15
environment:
POSTGRES_DB: macau_library
volumes:
- pg_data:/var/lib/postgresql/data
backend-api:
build: ./backend
depends_on:
- deepseek-llm
- vector-db
- postgres
Database Schema (Core Tables):
-- Library tabular data (System A)
CREATE TABLE library_tables (
id SERIAL PRIMARY KEY,
table_name VARCHAR(255) NOT NULL,
table_type VARCHAR(50), -- circulation, privileges, loans, resources
original_data JSONB,
normalized_data JSONB,
embedding_id VARCHAR(255),
created_at TIMESTAMP DEFAULT NOW(),
updated_at TIMESTAMP DEFAULT NOW()
);
-- Academic materials (System B)
CREATE TABLE academic_materials (
id SERIAL PRIMARY KEY,
title VARCHAR(500) NOT NULL,
material_type VARCHAR(50), -- thesis, book, paper
authors TEXT[],
abstract TEXT,
keywords TEXT[],
metadata JSONB,
embedding_id VARCHAR(255),
created_at TIMESTAMP DEFAULT NOW()
);
-- Query logs for improvement
CREATE TABLE query_logs (
id SERIAL PRIMARY KEY,
user_id VARCHAR(255),
query_text TEXT,
system_type CHAR(1), -- A or B
response TEXT,
feedback_score INTEGER,
created_at TIMESTAMP DEFAULT NOW()
);
Acceptance Criteria
- All Docker containers start successfully
- DeepSeek model responds to test prompts < 5 seconds
- SYVA admin panel accessible
- Database migrations complete
Sprint 3-4: System A - Tabular Data Pipeline (12 days)
Deliverables
| Task | Owner | Days | Priority |
|---|---|---|---|
| Table extraction & normalization scripts | Backend | 3 | P0 |
| Table preprocessing pipeline | ML Engineer | 2 | P0 |
| Hybrid reasoning module | ML Engineer | 3 | P0 |
| Query-to-structure translator | Backend | 2 | P1 |
| Table CRUD API endpoints | Backend | 2 | P1 |
Technical Implementation
Table Preprocessing Pipeline:
# /backend/services/table_processor.py
class TableProcessor:
"""
Normalizes and structures library tabular data for LLM consumption
"""
def normalize_table(self, raw_table: dict) -> dict:
"""
Steps:
1. Flatten multi-level headers
2. Separate mixed data types into columns
3. Standardize column names (snake_case)
4. Handle merged cells
5. Convert to consistent format
"""
pass
def generate_table_description(self, table: dict) -> str:
"""
Creates natural language description of table structure
for LLM context injection
"""
pass
def create_lookup_index(self, table: dict) -> dict:
"""
Creates inverted index for fast lookup operations
"""
pass
Hybrid Reasoning Architecture:
# /backend/services/hybrid_reasoner.py
class HybridReasoner:
"""
Combines LLM understanding with symbolic reasoning
"""
def process_query(self, query: str, tables: list) -> dict:
# Step 1: Intent classification
intent = self.classify_intent(query)
# Step 2: Entity extraction
entities = self.extract_entities(query)
# Step 3: Symbolic lookup (for exact matches)
if intent in ['lookup', 'comparison']:
symbolic_result = self.symbolic_lookup(entities, tables)
# Step 4: LLM reasoning (for complex queries)
llm_context = self.build_context(query, tables, symbolic_result)
llm_response = self.llm_inference(llm_context)
# Step 5: Merge and validate
return self.merge_results(symbolic_result, llm_response)
API Endpoints (System A)
POST /api/v1/tables # Upload/create table
GET /api/v1/tables # List all tables
GET /api/v1/tables/{id} # Get specific table
PUT /api/v1/tables/{id} # Update table
DELETE /api/v1/tables/{id} # Delete table
POST /api/v1/tables/query # Natural language query
POST /api/v1/tables/batch-import # Bulk import tables
Test Cases
# Example test queries for validation
test_queries = [
"What is the loan period for PhD students?",
"Can undergraduate students borrow journals?",
"Compare borrowing limits between staff and students",
"How many items can a visiting scholar borrow?",
"What are the overdue penalties for reserve materials?"
]
Sprint 5-6: System B - RAG Recommendation Engine (12 days)
Deliverables
| Task | Owner | Days | Priority |
|---|---|---|---|
| Document ingestion pipeline | Backend | 2 | P0 |
| Embedding generation service | ML Engineer | 2 | P0 |
| Vector similarity search | Backend | 2 | P0 |
| RAG retrieval module | ML Engineer | 3 | P0 |
| Recommendation API endpoints | Backend | 2 | P1 |
| Explanation generator | ML Engineer | 1 | P2 |
Technical Implementation
Document Ingestion Pipeline:
# /backend/services/document_ingestion.py
class DocumentIngestion:
"""
Processes up to 300 documents for recommendation system
"""
def __init__(self):
self.embedding_model = SentenceTransformer('all-MiniLM-L6-v2')
self.vector_db = QdrantClient()
def process_document(self, doc: dict) -> dict:
# 1. Extract metadata
metadata = self.extract_metadata(doc)
# 2. Clean and chunk content
chunks = self.chunk_content(doc['content'], chunk_size=512)
# 3. Generate embeddings
embeddings = self.embedding_model.encode(chunks)
# 4. Store in vector DB
self.vector_db.upsert(
collection_name="academic_materials",
points=[
PointStruct(
id=f"{doc['id']}_{i}",
vector=emb.tolist(),
payload={"doc_id": doc['id'], "chunk": chunk, **metadata}
)
for i, (emb, chunk) in enumerate(zip(embeddings, chunks))
]
)
def batch_process(self, documents: list):
"""Process up to 300 documents"""
for doc in tqdm(documents):
self.process_document(doc)
RAG Pipeline:
# /backend/services/rag_engine.py
class RAGEngine:
"""
Retrieval-Augmented Generation for recommendations
"""
def __init__(self):
self.retriever = VectorRetriever()
self.reranker = CrossEncoderReranker()
self.llm = DeepSeekClient()
def recommend(self, query: str, top_k: int = 5) -> list:
# 1. Initial retrieval (top 20)
candidates = self.retriever.search(query, k=20)
# 2. Re-rank with cross-encoder
reranked = self.reranker.rerank(query, candidates, k=top_k)
# 3. Generate recommendations with explanations
prompt = self.build_recommendation_prompt(query, reranked)
response = self.llm.generate(prompt)
return self.parse_recommendations(response)
def build_recommendation_prompt(self, query: str, docs: list) -> str:
return f"""Based on the user's research interest: "{query}"
Recommend the most relevant academic materials from the following candidates:
{self.format_candidates(docs)}
For each recommendation:
1. Explain WHY it's relevant to the query
2. Highlight key topics/methods that match
3. Suggest how it might be useful
Provide recommendations in order of relevance."""
API Endpoints (System B)
POST /api/v1/materials # Add material
GET /api/v1/materials # List materials
POST /api/v1/materials/bulk-import # Bulk import (max 300)
POST /api/v1/recommend # Get recommendations
POST /api/v1/recommend/explain # Get recommendation with explanation
GET /api/v1/recommend/similar/{id} # Find similar materials
Sprint 7-8: Integration & Chatbot Backend (10 days)
Deliverables
| Task | Owner | Days | Priority |
|---|---|---|---|
| Unified query router | Backend | 2 | P0 |
| Conversation state management | Backend | 2 | P0 |
| WebSocket real-time chat | Backend | 2 | P0 |
| User authentication integration | Backend | 2 | P1 |
| Access control implementation | Backend | 2 | P1 |
Conversation Flow
# /backend/services/conversation_manager.py
class ConversationManager:
"""
Manages multi-turn conversations with context
"""
def __init__(self):
self.sessions = {} # In-memory or Redis
def process_message(self, session_id: str, message: str) -> dict:
# Get or create session
session = self.get_session(session_id)
# Classify intent and route
intent = self.classify_intent(message, session.context)
if intent['system'] == 'A':
response = self.table_inquiry.process(message, session.context)
elif intent['system'] == 'B':
response = self.recommendation.process(message, session.context)
else:
response = self.general_response(message)
# Update context
session.add_turn(message, response)
return {
"response": response,
"follow_up_suggestions": self.generate_suggestions(session),
"sources": response.get("sources", [])
}
Sprint 9-10: Testing, Fine-tuning & Deployment (11 days)
Deliverables
| Task | Owner | Days | Priority |
|---|---|---|---|
| Unit & integration tests | All | 2 | P0 |
| Performance benchmarking | DevOps | 1 | P0 |
| LLM fine-tuning on library data | ML Engineer | 3 | P1 |
| UAT environment setup | DevOps | 1 | P0 |
| Production deployment | DevOps | 2 | P0 |
| Documentation | All | 2 | P1 |
Benchmark Dataset Structure
{
"system_a_benchmarks": [
{
"query": "What's the loan period for faculty members?",
"expected_answer": "28 days",
"table_source": "circulation_policies",
"difficulty": "easy"
},
{
"query": "Compare renewal limits between postgrad and undergrad",
"expected_answer": "Postgrad: 3 renewals, Undergrad: 2 renewals",
"table_source": "user_privileges",
"difficulty": "medium"
}
],
"system_b_benchmarks": [
{
"query": "Papers about machine learning in education",
"expected_materials": ["doc_001", "doc_015", "doc_042"],
"relevance_threshold": 0.7
}
]
}
PART 2: UX TEAM PLAN
Design Sprints Aligned with Development
UX Sprint 1: Research & Information Architecture (Week 1-2)
Deliverables
| Task | Owner | Days | Priority |
|---|---|---|---|
| User research (library staff interviews) | UX Researcher | 2 | P0 |
| Persona development | UX Designer | 1 | P0 |
| User journey mapping | UX Designer | 1.5 | P0 |
| Information architecture | UX Designer | 1.5 | P0 |
| Competitive analysis | UX Researcher | 1 | P1 |
Key Personas
Persona 1: Student Researcher (Sarah)
- Year 3 undergraduate
- Needs: Quick answers about borrowing policies, material recommendations for thesis
- Pain points: Complex library website, unclear privilege information
- Tech comfort: High
Persona 2: Library Staff (Michael)
- Circulation desk supervisor
- Needs: Manage tables, answer patron questions quickly
- Pain points: Updating policy tables, explaining rules to students
- Tech comfort: Medium
Persona 3: Faculty Member (Dr. Wong)
- Associate Professor
- Needs: Find relevant research papers, understand borrowing privileges
- Pain points: Time-consuming literature search
- Tech comfort: Medium-High
User Journey Map (Student Research Flow)
┌─────────┐ ┌──────────┐ ┌─────────────┐ ┌────────────┐ ┌─────────┐
│ Trigger │ → │ Discover │ → │ Query │ → │ Evaluate │ → │ Action │
│ │ │ │ │ │ │ │ │ │
│ Need to │ │ Find │ │ Ask natural │ │ Review │ │ Borrow/ │
│ find │ │ chatbot │ │ language │ │ results & │ │ Access │
│ papers │ │ on site │ │ question │ │ explanations│ │ materials│
└─────────┘ └──────────┘ └─────────────┘ └────────────┘ └─────────┘
│ │ │ │ │
▼ ▼ ▼ ▼ ▼
Emotions: Curious/ Hopeful/ Satisfied/ Accomplished
Frustrated Interested Uncertain Informed
UX Sprint 2: Wireframing & Prototyping (Week 3-4)
Deliverables
| Task | Owner | Days | Priority |
|---|---|---|---|
| Chatbot wireframes (mobile + desktop) | UX Designer | 2 | P0 |
| Admin CMS wireframes | UX Designer | 2 | P0 |
| Interactive prototype | UX Designer | 2 | P0 |
| Design system foundation | UI Designer | 2 | P1 |
Chatbot Interface Wireframes
Draft Mockups (Stitch): View Interactive Prototype
Main Chat View (Desktop)
┌────────────────────────────────────────────────────────────────────┐
│ ┌────────────────────────────────────────────────────────────────┐ │
│ │ 🎓 Macau University Library Assistant [−][□][×] │ │
│ └────────────────────────────────────────────────────────────────┘ │
│ │
│ ┌────────────────────────────────────────────────────────────────┐ │
│ │ │ │
│ │ 👋 Hello! I can help you with: │ │
│ │ │ │
│ │ 📊 Library policies & borrowing information │ │
│ │ 📚 Academic material recommendations │ │
│ │ │ │
│ │ What would you like to know today? │ │
│ │ │ │
│ └────────────────────────────────────────────────────────────────┘ │
│ │
│ ┌────────────────────────────────────────────────────────────────┐ │
│ │ 💬 User: How many books can I borrow as a PhD student? │ │
│ └────────────────────────────────────────────────────────────────┘ │
│ │
│ ┌────────────────────────────────────────────────────────────────┐ │
│ │ 🤖 Assistant: │ │
│ │ │ │
│ │ As a PhD student, you can borrow up to **30 items** at a │ │
│ │ time with a loan period of **28 days**. │ │
│ │ │ │
│ │ ┌──────────────────────────────────────────┐ │ │
│ │ │ 📋 Source: User Privileges Table │ │ │
│ │ │ Updated: Jan 2025 │ │ │
│ │ └──────────────────────────────────────────┘ │ │
│ │ │ │
│ │ Would you like to know about renewals or reserves? │ │
│ │ │ │
│ │ [Renewals] [Reserves] [Other privileges] │ │
│ │ │ │
│ └────────────────────────────────────────────────────────────────┘ │
│ │
│ ┌────────────────────────────────────────────────────────────────┐ │
│ │ [Type your question here...] [Send] │ │
│ └────────────────────────────────────────────────────────────────┘ │
└────────────────────────────────────────────────────────────────────┘
Recommendation Response View
┌────────────────────────────────────────────────────────────────────┐
│ 💬 User: Find papers about deep learning in medical imaging │
└────────────────────────────────────────────────────────────────────┘
┌────────────────────────────────────────────────────────────────────┐
│ 🤖 I found 5 relevant materials for your research: │
│ │
│ ┌──────────────────────────────────────────────────────────────┐ │
│ │ 📄 1. "Deep Learning for Medical Image Analysis" │ │
│ │ Authors: Chen et al. (2023) │ │
│ │ Type: Thesis | Relevance: ★★★★★ │ │
│ │ │ │
│ │ WHY RELEVANT: This thesis directly addresses CNN │ │
│ │ architectures for radiology imaging, matching your │ │
│ │ interest in medical applications. │ │
│ │ │ │
│ │ [View Details] [Add to Reading List] │ │
│ └──────────────────────────────────────────────────────────────┘ │
│ │
│ ┌──────────────────────────────────────────────────────────────┐ │
│ │ 📄 2. "Transfer Learning in Diagnostic Systems" │ │
│ │ Authors: Wong & Liu (2024) │ │
│ │ Type: Journal Article | Relevance: ★★★★☆ │ │
│ │ ... │ │
│ └──────────────────────────────────────────────────────────────┘ │
│ │
│ [Load more results] │
│ │
│ Related questions: │
│ • "Narrow to COVID-19 imaging" │
│ • "Include papers from 2023 only" │
│ • "Show similar theses" │
└────────────────────────────────────────────────────────────────────┘
Admin CMS - Table Management
┌────────────────────────────────────────────────────────────────────┐
│ SYVA Admin │ Tables │ Documents │ Users │ Analytics │ Settings │
├────────────────────────────────────────────────────────────────────┤
│ │
│ 📊 Library Tables [+ Add New Table] │
│ │
│ ┌──────────────────────────────────────────────────────────────┐ │
│ │ Search tables... [Filter ▼] [Sort ▼]│ │
│ └──────────────────────────────────────────────────────────────┘ │
│ │
│ ┌─────────────────────────────────────────────────────────────┐ │
│ │ Table Name │ Type │ Last Updated │ Actions │ │
│ ├─────────────────────┼─────────────┼──────────────┼──────────┤ │
│ │ Circulation Policy │ Policies │ Jan 15, 2025 │ [✏️][👁️][🗑️]│ │
│ │ User Privileges │ Privileges │ Jan 10, 2025 │ [✏️][👁️][🗑️]│ │
│ │ Loan Periods │ Loans │ Jan 8, 2025 │ [✏️][👁️][🗑️]│ │
│ │ Resource Limits │ Resources │ Dec 28, 2024 │ [✏️][👁️][🗑️]│ │
│ └─────────────────────┴─────────────┴──────────────┴──────────┘ │
│ │
│ Showing 4 of 10 tables [< 1 2 3 >] │
└────────────────────────────────────────────────────────────────────┘
UX Sprint 3: Visual Design & Components (Week 5-6)
Deliverables
| Task | Owner | Days | Priority |
|---|---|---|---|
| Visual design system | UI Designer | 3 | P0 |
| Chat UI components | UI Designer | 2 | P0 |
| Admin UI components | UI Designer | 2 | P1 |
| Responsive design specs | UI Designer | 1 | P0 |
| Design handoff documentation | UI Designer | 1 | P1 |
Design System Specifications
Color Palette
Primary: #1E3A5F (Deep Navy - Library/Academic)
Secondary: #2E7D32 (Forest Green - Success)
Accent: #F57C00 (Amber - Highlights)
Background: #F5F5F5 (Light Gray)
Surface: #FFFFFF (White)
Text: #212121 (Near Black)
Text Light: #757575 (Gray)
Error: #D32F2F (Red)
Typography
Headings: Inter (Bold)
Body: Inter (Regular)
Code/Tables: Roboto Mono
H1: 24px / 32px line-height
H2: 20px / 28px line-height
Body: 16px / 24px line-height
Caption: 14px / 20px line-height
Chat Components
- Message bubble (user): Right-aligned, Primary color
- Message bubble (assistant): Left-aligned, Surface color with border
- Quick action chips: Rounded, Accent color outline
- Source citation cards: Surface with left border (Primary)
- Loading indicator: Typing animation (3 dots)
UX Sprint 4: Usability Testing & Refinement (Week 7-8)
Deliverables
| Task | Owner | Days | Priority |
|---|---|---|---|
| Usability test planning | UX Researcher | 1 | P0 |
| User testing sessions (5 users) | UX Researcher | 2 | P0 |
| Analysis & recommendations | UX Researcher | 1 | P0 |
| Design iterations | UI Designer | 2 | P0 |
| Final design specs | UI Designer | 1 | P0 |
Usability Test Script (Key Tasks)
Task 1: Policy Inquiry
“You’re a new PhD student and want to know how many books you can borrow. Use the chatbot to find out.”
Success Criteria:
- User finds chatbot within 30 seconds
- User formulates question naturally
- User understands the response
- User can find source information
Task 2: Material Recommendation
“You’re writing a thesis on AI in healthcare. Ask the assistant to recommend relevant papers.”
Success Criteria:
- User successfully requests recommendations
- User understands why materials are recommended
- User can take action on recommendations (view details, save)
Task 3: Follow-up Question
“After getting the borrowing limit, ask about renewals without repeating context.”
Success Criteria:
- System maintains conversation context
- User feels the interaction is natural
PART 3: INTEGRATED TIMELINE
Week | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 |
------|----|----|----|----|----|----|----|----|----|----|
DEV |███████ Infrastructure ██████|
| |███████ System A - Tables ██████|
| |███████ System B - RAG █████████|
| |█████ Integration █████|
| |██ Test/Deploy ███|
------|----|----|----|----|----|----|----|----|----|----|
UX |██ Research ██|
| |███ Wireframes ███|
| |███ Visual Design ███|
| |██ Testing ██|
------|----|----|----|----|----|----|----|----|----|----|
PART 4: KEY DELIVERABLES CHECKLIST
Development Deliverables
- Docker deployment package
- DeepSeek 8B model configuration
- SYVA platform setup
- Table preprocessing pipeline
- RAG recommendation engine
- REST API documentation
- WebSocket chat service
- Admin API for CRUD operations
- User authentication module
- Unit & integration tests
- Performance benchmark results
- Deployment runbook
UX Deliverables
- User personas (3)
- User journey maps (3 flows)
- Information architecture diagram
- Wireframes (mobile + desktop)
- Interactive prototype
- Design system documentation
- Component library
- Usability test report
- Final design specifications
- Developer handoff package
PART 5: RISK REGISTER
| Risk | Impact | Likelihood | Mitigation |
|---|---|---|---|
| LLM accuracy on complex tables | High | Medium | Hybrid reasoning + fine-tuning |
| Document volume exceeds 300 | Medium | Low | Batch processing, pagination |
| Integration with SYVA delays | High | Medium | Early setup, backup plan |
| User adoption resistance | Medium | Medium | Training sessions, gradual rollout |
| GPU memory constraints | High | Low | Model quantization, batching |
PART 6: SUCCESS METRICS
System A (Tabular Inquiry)
- Query accuracy: ≥ 90%
- Response time: < 3 seconds
- User satisfaction: ≥ 4.0/5.0
System B (Recommendations)
- Recommendation relevance: ≥ 80% (user feedback)
- Click-through rate: ≥ 30%
- Return user rate: ≥ 50%
Overall
- UAT pass rate: 100% critical, 90% all
- System uptime: ≥ 99%
- Training completion: 100% staff
APPENDIX A: Tech Stack Summary
| Component | Technology |
|---|---|
| LLM | DeepSeek-R1-Distill-Llama 8B |
| Vector DB | Qdrant or ChromaDB |
| Database | PostgreSQL 15 |
| Backend | Python 3.11 + FastAPI |
| Frontend | React 18 + TypeScript |
| CMS | SYVA Platform |
| Deployment | Docker + Docker Compose |
| Embedding | sentence-transformers |
APPENDIX B: API Contract Summary
openapi: 3.0.0
info:
title: Macau Library LLM API
version: 1.0.0
paths:
/api/v1/chat:
post:
summary: Send chat message
requestBody:
content:
application/json:
schema:
type: object
properties:
session_id: string
message: string
responses:
200:
description: Chat response
/api/v1/tables/query:
post:
summary: Query tabular data
/api/v1/recommend:
post:
summary: Get material recommendations
Document Version: 1.0
Created: Based on Macau U Proposal v_Revised_06022025
Status: Ready for Team Review
**