1Problems với hardcoded prompts:
2- Khó update/modify
3- Không A/B test được
4- Duplicate code
5- Khó maintain nhiều AI features

1Data Type: Prompt
2- name (text, unique)
3- system_prompt (text)
4- user_template (text)
5- model (text)
6- temperature (number)
7- max_tokens (number)
8- is_active (yes/no)
9- version (number)

1App Settings:
2- prompt_summarize
3- prompt_translate
4- prompt_chat_system
5- prompt_content_gen

1Store prompts in Bubble's Backend:
2- Scheduled API calls
3- Reusable API workflows
4- Centralized management

1┌────────────────────────────────────────┐
2│  Prompt Manager              [+ New]   │
3├────────────────────────────────────────┤
4│ ┌──────────────────────────────────┐   │
5│ │ content_generator          [Edit]│   │
6│ │ v2 | gpt-4o-mini | Active       │   │
7│ └──────────────────────────────────┘   │
8│ ┌──────────────────────────────────┐   │
9│ │ summarizer                 [Edit]│   │
10│ │ v1 | gpt-4o-mini | Active       │   │
11│ └──────────────────────────────────┘   │
12└────────────────────────────────────────┘

1Form fields:
21. Name (unique identifier)
32. System Prompt (multiline)
43. User Template (with {variables})
54. Model Selection
65. Temperature slider
76. Max Tokens
87. Test Panel

1User Template:
2"Write a {content_type} about {topic} in {tone} tone.
3Target audience: {audience}
4Length: {length} words"
5 
6Variables:
7- {content_type}: blog, email, ad
8- {topic}: user input
9- {tone}: professional, casual
10- {audience}: dropdown
11- {length}: number input

1In Bubble:
2 
31. Get prompt template from database
42. Use "Find & Replace" plugin
5   - Find: {topic}
6   - Replace with: Input Topic's value
73. Repeat for each variable
84. Send to API

1Each prompt change:
21. Keep old version (mark inactive)
32. Create new version
43. Track performance per version
5 
6Data structure:
7Prompt:
8- name
9- version (number)
10- is_active (yes/no)
11- content
12- created_date
13- performance_score

1Setup:
2- PromptA (50% traffic)
3- PromptB (50% traffic)
4 
5Random selection:
61. Generate random number
72. If < 0.5 → PromptA
83. Else → PromptB
94. Log which was used
105. Track results

1System prompt enhancement:
2 
3"You are assisting {user_name}.
4Their account type: {account_type}
5Previous interactions: {interaction_count}
6Preferences: {user_preferences}"

1Add app-specific context:
2 
3"Current app state: {page_context}
4User's recent actions: {recent_actions}
5Related content: {relevant_data}"

1❌ Bad:
2"Write something about AI"
3 
4✅ Good:
5"Write a 200-word introduction to AI for beginners.
6Use simple language, avoid jargon.
7Include one real-world example.
8End with a question to engage readers."

1In Bubble - Backend Workflows:
2 
3Workflow: generate_content
4Parameters:
5- prompt_name (text)
6- variables (text/JSON)
7- user_id (User)
8 
9Steps:
101. Get Prompt where name = prompt_name
112. Replace variables in template
123. Call OpenAI API
134. Log usage
145. Return response

1Workflow:
21. Button clicked
32. Schedule API workflow: generate_content
4   - prompt_name: "blog_intro"
5   - variables: JSON with user inputs
63. Wait for response
74. Display result

1Cache strategy:
21. Hash input (prompt + variables)
32. Check cache database
43. If hit → return cached response
54. If miss → call API, save to cache
6 
7Cache Data Type:
8- input_hash
9- response
10- created_date
11- hit_count

1Per-user limits:
2- Free: 10 requests/day
3- Pro: 100 requests/day
4- Enterprise: Unlimited
5 
6Track in database:
7- user_id
8- request_count
9- last_reset_date

1Log each AI call:
2 
3AILog:
4- prompt_used
5- user_input
6- full_request
7- response
8- response_time
9- tokens_used
10- error (if any)

1Build admin test interface:
21. Select prompt
32. Enter test variables
43. Run test
54. View full request/response
65. Adjust and iterate