- BottleGrid: Implement blurred backdrop effect for bottle cards - Cascade OCR: TextDetector → RegEx → Fuzzy Match → window.ai pipeline - Smart Scan: Native OCR for Android, Live Text fallback for iOS - OCR Dashboard: Admin page at /admin/ocr-logs with stats and scan history - Features: Add feature flags in src/config/features.ts - SQL: Add ocr_logs table migration - Services: Update analyze-bottle to use OpenRouter, add save-ocr-log
275 lines
9.7 KiB
TypeScript
275 lines
9.7 KiB
TypeScript
'use server';
|
|
|
|
import { BottleMetadataSchema, BottleMetadata } from '@/types/whisky';
|
|
import { createClient } from '@/lib/supabase/server';
|
|
import { trackApiUsage } from '@/services/track-api-usage';
|
|
import { checkCreditBalance, deductCredits } from '@/services/credit-service';
|
|
import { getAIProvider, getOpenRouterClient, OPENROUTER_VISION_MODEL, OPENROUTER_PROVIDER_PREFERENCES } from '@/lib/openrouter';
|
|
import { normalizeWhiskyData } from '@/lib/distillery-matcher';
|
|
import { formatWhiskyName } from '@/utils/formatWhiskyName';
|
|
import { createHash } from 'crypto';
|
|
|
|
const VISION_PROMPT = `ROLE: Senior Whisky Database Curator.
|
|
|
|
OBJECTIVE: Extract metadata from the bottle image.
|
|
CRITICAL: You must CONSTRUCT the product name from its parts, not just read lines.
|
|
|
|
INPUT IMAGE ANALYSIS (Mental Steps):
|
|
1. FIND DISTILLERY: Look for the most prominent brand name (e.g. "Annandale").
|
|
2. FIND AGE: Scan the center for "Aged X Years" or large numbers. (In this image: Look for a large "10" in the center).
|
|
3. FIND VINTAGE: Look for small script/cursive text like "Distilled 2011".
|
|
4. FIND CASK TYPE: Look for specific maturation terms like 'Sherry', 'Bourbon', 'Port', 'Oloroso', 'PX', 'Madeira', 'Rum', 'Hogshead', 'Butt', 'Barrel', or 'Finish'. Extract ONLY this phrase into this field (e.g., 'Oloroso Cask Matured') cask_type. Do not leave null if these words are visible.
|
|
5. FIND SERIES: Look at the top logo (e.g. "Cadenhead's Natural Strength").
|
|
|
|
COMPOSITION RULES (How to fill the 'name' field):
|
|
- DO NOT just write "Single Malt".
|
|
- Format: "[Age/Vintage] [Series] [Cask Info]"
|
|
- Example: "10 Year Old Cadenhead's Natural Strength Oloroso Matured"
|
|
|
|
OUTPUT SCHEMA (Strict JSON):
|
|
{
|
|
"name": "string (The constructed full name based on rules above)",
|
|
"distillery": "string",
|
|
"bottler": "string",
|
|
"series": "string (e.g. Natural Strength)",
|
|
"abv": number,
|
|
"age": numberOrNull,
|
|
"vintage": "stringOrNull",
|
|
"cask_type": "stringOrNull",
|
|
"distilled_at": "stringOrNull",
|
|
"bottled_at": "stringOrNull",
|
|
"batch_info": "stringOrNull",
|
|
"is_whisky": true,
|
|
"confidence": number
|
|
}`;
|
|
|
|
export interface ScannerResult {
|
|
success: boolean;
|
|
data?: BottleMetadata;
|
|
error?: string;
|
|
provider?: 'openrouter';
|
|
perf?: {
|
|
imagePrep?: number;
|
|
apiCall: number;
|
|
total: number;
|
|
cacheHit?: boolean;
|
|
apiDuration?: number;
|
|
parsing?: number;
|
|
parseDuration?: number;
|
|
uploadSize?: number;
|
|
cacheCheck?: number;
|
|
encoding?: number;
|
|
modelInit?: number;
|
|
validation?: number;
|
|
dbOps?: number;
|
|
};
|
|
raw?: any;
|
|
search_string?: string;
|
|
}
|
|
|
|
/**
|
|
* Compatibility wrapper for older call sites that use an object/FormData input.
|
|
*/
|
|
export async function scanLabel(input: any): Promise<ScannerResult> {
|
|
const getValue = (obj: any, key: string): any => {
|
|
if (obj && typeof obj.get === 'function') return obj.get(key);
|
|
if (obj && typeof obj[key] !== 'undefined') return obj[key];
|
|
return null;
|
|
};
|
|
|
|
const file = getValue(input, 'file');
|
|
if (file && file instanceof File) {
|
|
// Handle file input (e.g. from FormData)
|
|
const arrayBuffer = await file.arrayBuffer();
|
|
const base64 = Buffer.from(arrayBuffer).toString('base64');
|
|
return analyzeBottleLabel(`data:${file.type};base64,${base64}`);
|
|
}
|
|
|
|
if (typeof input === 'string') {
|
|
return analyzeBottleLabel(input);
|
|
}
|
|
|
|
return { success: false, error: 'Ungültiges Eingabeformat.' };
|
|
}
|
|
|
|
/**
|
|
* Unified action for analyzing a whisky bottle label.
|
|
* Replaces redundant gemini-vision.ts, scan-label.ts, and analyze-bottle.ts.
|
|
*/
|
|
export async function analyzeBottleLabel(imageBase64: string): Promise<ScannerResult> {
|
|
const startTotal = performance.now();
|
|
const provider = getAIProvider();
|
|
|
|
if (!imageBase64 || imageBase64.length < 100) {
|
|
return { success: false, error: 'Ungültige Bilddaten.' };
|
|
}
|
|
|
|
try {
|
|
// 1. Auth & Credit Check
|
|
const supabase = await createClient();
|
|
const { data: { user } } = await supabase.auth.getUser();
|
|
if (!user) return { success: false, error: 'Nicht autorisiert.' };
|
|
|
|
const creditCheck = await checkCreditBalance(user.id, 'gemini_ai');
|
|
if (!creditCheck.allowed) {
|
|
return {
|
|
success: false,
|
|
error: `Nicht genügend Credits. Benötigt: ${creditCheck.cost}, Verfügbar: ${creditCheck.balance}.`
|
|
};
|
|
}
|
|
|
|
// 2. Extract base64 and hash for caching
|
|
let base64Data = imageBase64;
|
|
let mimeType = 'image/webp';
|
|
if (imageBase64.startsWith('data:')) {
|
|
const matches = imageBase64.match(/^data:([^;]+);base64,(.+)$/);
|
|
if (matches) {
|
|
mimeType = matches[1];
|
|
base64Data = matches[2];
|
|
}
|
|
}
|
|
|
|
const buffer = Buffer.from(base64Data, 'base64');
|
|
const imageHash = createHash('sha256').update(buffer).digest('hex');
|
|
|
|
// 3. Cache Check
|
|
const { data: cachedResult } = await supabase
|
|
.from('vision_cache')
|
|
.select('result')
|
|
.eq('hash', imageHash)
|
|
.maybeSingle();
|
|
|
|
if (cachedResult) {
|
|
console.log(`[Scanner] Cache HIT for hash: ${imageHash.slice(0, 8)}...`);
|
|
return {
|
|
success: true,
|
|
data: cachedResult.result as BottleMetadata,
|
|
perf: {
|
|
apiCall: 0,
|
|
total: performance.now() - startTotal,
|
|
cacheHit: true
|
|
}
|
|
};
|
|
}
|
|
|
|
// 4. AI Analysis with retry logic for OpenRouter
|
|
console.log(`[Scanner] Using provider: ${provider}`);
|
|
let aiResult: { data: any; apiTime: number; responseText: string };
|
|
|
|
const client = getOpenRouterClient();
|
|
const startApi = performance.now();
|
|
const maxRetries = 3;
|
|
let lastError: any = null;
|
|
let response: any = null;
|
|
|
|
for (let attempt = 1; attempt <= maxRetries; attempt++) {
|
|
try {
|
|
response = await client.chat.completions.create({
|
|
model: OPENROUTER_VISION_MODEL,
|
|
messages: [
|
|
{
|
|
role: 'user',
|
|
content: [
|
|
{ type: 'image_url', image_url: { url: `data:${mimeType};base64,${base64Data}` } },
|
|
{ type: 'text', text: VISION_PROMPT },
|
|
],
|
|
},
|
|
],
|
|
temperature: 0.1,
|
|
max_tokens: 1024,
|
|
// @ts-ignore
|
|
provider: OPENROUTER_PROVIDER_PREFERENCES,
|
|
});
|
|
break; // Success!
|
|
} catch (err: any) {
|
|
lastError = err;
|
|
if (err.status === 429 && attempt < maxRetries) {
|
|
const delay = Math.pow(2, attempt) * 1000;
|
|
console.warn(`[Scanner] Rate limited (429). Retrying in ${delay}ms...`);
|
|
await new Promise(resolve => setTimeout(resolve, delay));
|
|
continue;
|
|
}
|
|
throw err;
|
|
}
|
|
}
|
|
|
|
if (!response) throw lastError || new Error('OpenRouter response failed after retries');
|
|
|
|
const content = response.choices[0]?.message?.content || '{}';
|
|
let jsonStr = content;
|
|
const jsonMatch = content.match(/```(?:json)?\s*([\s\S]*?)```/) || content.match(/\{[\s\S]*\}/);
|
|
if (jsonMatch) {
|
|
jsonStr = jsonMatch[jsonMatch.length - 1].trim();
|
|
}
|
|
aiResult = {
|
|
data: JSON.parse(jsonStr),
|
|
apiTime: performance.now() - startApi,
|
|
responseText: content
|
|
};
|
|
|
|
// 6. Name Composition & Normalization
|
|
// Use standardized helper to construct the perfect name
|
|
console.log(`[Uncleaned Data]: ${JSON.stringify(aiResult.data)}`);
|
|
|
|
const d = aiResult.data;
|
|
const constructedName = formatWhiskyName({
|
|
distillery: d.distillery || '',
|
|
bottler: d.bottler,
|
|
series: d.series,
|
|
age: d.age,
|
|
vintage: d.vintage,
|
|
cask_type: d.cask_type
|
|
}) || d.name;
|
|
|
|
// Validation & Normalization
|
|
const validatedData = BottleMetadataSchema.parse({
|
|
...d,
|
|
name: constructedName
|
|
});
|
|
|
|
const normalized = normalizeWhiskyData(
|
|
validatedData.name || '',
|
|
validatedData.distillery || ''
|
|
);
|
|
|
|
const finalData = {
|
|
...validatedData,
|
|
name: normalized.name || validatedData.name,
|
|
distillery: normalized.distillery || validatedData.distillery,
|
|
};
|
|
|
|
// 7. Success Tracking & Caching
|
|
await trackApiUsage({
|
|
userId: user.id,
|
|
apiType: 'gemini_ai',
|
|
endpoint: `analyzeBottleLabel_${provider}`,
|
|
success: true,
|
|
provider,
|
|
model: OPENROUTER_VISION_MODEL,
|
|
responseText: aiResult.responseText
|
|
});
|
|
await deductCredits(user.id, 'gemini_ai', `Scanner analysis (${provider})`);
|
|
|
|
await supabase.from('vision_cache').insert({ hash: imageHash, result: finalData });
|
|
|
|
return {
|
|
success: true,
|
|
data: finalData,
|
|
provider,
|
|
perf: {
|
|
apiCall: aiResult.apiTime,
|
|
apiDuration: aiResult.apiTime, // Compatibility
|
|
total: performance.now() - startTotal,
|
|
cacheHit: false
|
|
}
|
|
};
|
|
|
|
} catch (error: any) {
|
|
console.error(`[Scanner] Analysis failed:`, error);
|
|
return {
|
|
success: false,
|
|
error: error.message || 'Vision analysis failed.',
|
|
};
|
|
}
|
|
}
|