tech4lince/image_tagger
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

prompt 0.002

Browse Source
This commit is contained in:
laxman 2025-11-03 14:15:32 +05:30
parent e8301708a3
commit ab6d782212
1 changed files with 13 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
src/infrastructure/ai/ClaudeAIProvider.js
View File

@ -112,7 +112,7 @@ class ClaudeAIProvider extends IImageTaggingService {
You are an elite real estate property analyst AI with comprehensive expertise in architectural photography, interior design, property valuation, and regional market knowledge (Dubai, Mumbai, Singapore, London, NYC).
CORE MISSION: Generate exactly 30 precise, high-confidence tags from property images that drive listing performance and buyer engagement.
CORE MISSION: Generate 25-30 precise, high-confidence tags from property images that drive listing performance and buyer engagement. Quality over quantity - omit categories that are absent rather than adding placeholder tags.
## QUALITY STANDARDS
- Every tag must be visually verifiable (>92% expert accuracy)
@ -199,12 +199,15 @@ Adjust for image quality: poor (-0.10), excellent (+0.05), multiple confirmation
7. Luxury tags → neutral/monochrome color scheme more likely
## TAG DISTRIBUTION
**Exactly 30 tags total**:
**Target: 25-30 tags total** (maintain consistent quality):
- High prominence (3-4 each): Furnishing, Style, Features, Lighting
- Medium (2-3 each): Room Type, Flooring, Color Scheme, View
- Standard (2-3 each): Kitchen, Condition
Dynamic adjustment: Kitchen not visible → redistribute to Features/Furnishing/Style
Dynamic adjustment:
- Kitchen not visible → redistribute to Features/Furnishing/Style
- View not visible → redistribute to Features/Furnishing/Style
- Category completely absent → OMIT that category, do NOT add placeholder tags
## SUMMARY GENERATION
**Structure**: [Room Type] + [Key Feature 1] + [Key Feature 2] + [Style/Condition]
@ -218,9 +221,10 @@ Examples:
## EDGE CASES
- Unfurnished: tag unfurnished (0.95+), compensate with architectural Features/Flooring
- Multiple rooms: tag all visible rooms (3-4), add "open-plan" to Features
- No view: use "no exterior view" (0.85+) or redistribute tags
- Kitchen not visible: tag "not visible" (0.85+) or redistribute
- No view: OMIT View category entirely, redistribute tags to other categories
- Kitchen not visible: OMIT Kitchen category entirely, redistribute tags to other categories
- Poor quality: reduce all confidence scores by 0.10, use broader tags
- **CRITICAL**: Never add placeholder tags like "not visible" - if category is absent, omit it and maintain tag quality
## OUTPUT FORMAT
Return ONLY valid JSON (no markdown, explanations, or additional text):
@ -233,7 +237,7 @@ Return ONLY valid JSON (no markdown, explanations, or additional text):
{"category": "Furnishing", "value": "fully furnished", "confidence": 0.96},
{"category": "Furnishing", "value": "contemporary", "confidence": 0.93},
{"category": "Furnishing", "value": "luxury", "confidence": 0.90},
{"category": "Kitchen", "value": "not visible", "confidence": 0.85},
{"category": "Furnishing", "value": "modern", "confidence": 0.88},
{"category": "Flooring", "value": "marble", "confidence": 0.94},
{"category": "Flooring", "value": "polished", "confidence": 0.91},
{"category": "Flooring", "value": "light colored", "confidence": 0.87},
@ -297,10 +301,10 @@ Ready to analyze. Execute complete framework and return only JSON.`;
throw new AIServiceError('Invalid response: tags array is required');
}
// Validate exactly 30 tags as per enhanced prompt requirements
if (parsed.tags.length !== 30) {
// Validate tag count (25-30 range to allow for missing categories)
if (parsed.tags.length < 25 || parsed.tags.length > 30) {
this.logger.warn('Claude returned unexpected number of tags', {
expected: 30,
expectedRange: '25-30',
actual: parsed.tags.length
});
}