Analyzes your urban observation and produces a structured problem statement with clear scope, measurable impact, and a target user profile. It does NOT propose solutions — only frames the problem.

1. Create an LLM Agent node on the canvas
2. Name it "Problem Framer"
3. Set model to doubao-seed-2-0-lite-260215
4. Set temperature to 0.3 (low — consistent structured output)
5. Paste the system prompt from the node template
6. Define input variables: project_description (string), project_domain (string)
7. Define output variables: problem_statement (string), user_profile (string)

• Be specific in your project description — include location, time, and who is affected
• If the output is too broad, add geographic or demographic constraints to your input
• The node explicitly does NOT propose solutions — if it does, check your system prompt
• This node's outputs feed directly into the Data Stack Designer

Matches your project to specific Hong Kong government open data portals (DATA.GOV.HK, Transport Dept, EPD, etc.), identifies real-time vs. static data needs, and designs a data flow architecture from source to insight.

1. Create an LLM Agent node, name it "Data Stack Designer"
2. Model: doubao-seed-2-0-lite-260215, temperature 0.3
3. Paste the system prompt from the template
4. Define inputs: project_description, project_domain, problem_statement
5. Define output: data_sources (string)
6. Connect from Problem Framer (drag problem_statement edge)

• Prioritize FREE and publicly accessible data — the node is configured for this
• If government data doesn't cover your needs, the node suggests IoT sensors or crowdsourcing
• Be specific about your domain — transport, environment, energy, building, or public service
• The data architecture should be simple enough for a student prototype

Designs a 4-layer architecture (sensing → data → logic → presentation), maps data flow between layers, assesses smart capability level (perception → coordination → prediction → adaptation → value delivery), and connects to BIM/Digital Twin concepts.

1. Create an LLM Agent node, name it "System Architect"
2. Model: doubao-seed-2-0-lite-260215, temperature 0.3
3. Paste the system prompt
4. Define inputs: problem_statement, data_sources, project_domain
5. Define output: system_architecture (string)
6. Connect from Problem Framer and Data Stack Designer

• The 4 layers are: Sensing (data collection), Data (storage/structure), Logic (decisions/algorithms), Presentation (user interaction)
• BIM/Digital Twin connection is required — even a simple one counts
• Smart capability levels go from basic (perception) to advanced (value delivery)
• Keep the architecture implementable as a student prototype

Identifies specific AI capabilities needed (RAG, prediction, anomaly detection, NLP, computer vision), recommends implementation path (no-code vs code), designs AI Agent configuration, and defines workflow automation with triggers and feedback loops.

1. Create an LLM Agent node, name it "AI Composer"
2. Model: doubao-seed-2-0-lite-260215, temperature 0.3
3. Paste the system prompt
4. Define inputs: system_architecture, data_sources, problem_statement
5. Define output: ai_workflow_config (string)
6. Connect from System Architect, Data Stack Designer, and Problem Framer

• AI must serve a clear purpose — don't add AI just because you can
• Focus on 1-2 core AI capabilities that directly solve the problem
• The node recommends no-code/low-code vs code implementation
• Include specific system prompt and tool definitions for your Agent

Identifies all beneficiaries and their specific value, classifies value types (public/institutional/commercial), assesses 3 types of feasibility (technical, economic, policy), identifies risks and failure modes, and suggests an implementation roadmap.

1. Create an LLM Agent node, name it "Value Canvas"
2. Model: doubao-seed-2-0-lite-260215, temperature 0.3
3. Paste the system prompt
4. Define inputs: system_architecture, problem_statement, user_profile
5. Define output: value_proposition (string)
6. Connect from System Architect and Problem Framer

• Be specific about WHO benefits and HOW — avoid generic statements
• All 3 feasibility types must be addressed: technical, economic, policy
• The node references HK Smart City Blueprint 2.0 — know what it says
• Be honest about risks — the prompt explicitly asks for this

Synthesizes all outputs from the previous 5 nodes into a standardized Skill document (JSON format) that covers the 7 project requirements and can be installed on an AI Agent.

1. Create an LLM Agent node, name it "Skill Generator"
2. Model: doubao-seed-2-0-lite-260215, temperature 0.3
3. Paste the system prompt
4. Define ALL input variables from previous 5 nodes
5. Define output: skill_document (string)
6. Connect from ALL 5 previous nodes

• This is your final deliverable — make sure all 7 project requirements are covered
• The output must be valid JSON — test it with a JSON validator
• The document should be self-contained (an Agent can understand your project from it alone)
• Include Agent installation instructions so others can use your Skill
• This directly addresses the Final Project (50%) assessment

• 8-10 minute presentation covering: problem statement, data sources, initial system architecture
• Team of 3-4 students
• Q&A session (5 minutes)
• Submit slides (PDF) via Blackboard before class

1. Click "Run" on the canvas
2. Enter your project description and domain
3. Watch each node execute in sequence
4. Check the output of each node — does it make sense?
5. If a node fails, check its system prompt and input variables

• 12-15 minute presentation (Week 13)
• Cover: problem, data, architecture, AI, value, and demo
• Include your Skill document as a handout
• Practice with your team — timing matters

• Export your workflow as a Skill document
• Submit via Blackboard before the deadline
• Include: workflow screenshot, Skill JSON, and presentation slides