feat: Complete internationalization and code cleanup
## Translation Files - Add 11 new language files (es, de, pt, ru, zh, ja, ko, ar, hi, nl, pl) - Add 100+ missing translation keys across all 15 languages - New sections: notebook, pagination, ai.batchOrganization, ai.autoLabels - Update nav section with workspace, quickAccess, myLibrary keys ## Component Updates - Update 15+ components to use translation keys instead of hardcoded text - Components: notebook dialogs, sidebar, header, note-input, ghost-tags, etc. - Replace 80+ hardcoded English/French strings with t() calls - Ensure consistent UI across all supported languages ## Code Quality - Remove 77+ console.log statements from codebase - Clean up API routes, components, hooks, and services - Keep only essential error handling (no debugging logs) ## UI/UX Improvements - Update Keep logo to yellow post-it style (from-yellow-400 to-amber-500) - Change selection colors to #FEF3C6 (notebooks) and #EFB162 (nav items) - Make "+" button permanently visible in notebooks section - Fix grammar and syntax errors in multiple components ## Bug Fixes - Fix JSON syntax errors in it.json, nl.json, pl.json, zh.json - Fix syntax errors in notebook-suggestion-toast.tsx - Fix syntax errors in use-auto-tagging.ts - Fix syntax errors in paragraph-refactor.service.ts - Fix duplicate "fusion" section in nl.json 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com> Ou une version plus courte si vous préférez : feat(i18n): Add 15 languages, remove logs, update UI components - Create 11 new translation files (es, de, pt, ru, zh, ja, ko, ar, hi, nl, pl) - Add 100+ translation keys: notebook, pagination, AI features - Update 15+ components to use translations (80+ strings) - Remove 77+ console.log statements from codebase - Fix JSON syntax errors in 4 translation files - Fix component syntax errors (toast, hooks, services) - Update logo to yellow post-it style - Change selection colors (#FEF3C6, #EFB162) 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
This commit is contained in:
224
keep-notes/lib/ai/services/embedding.service.ts
Normal file
224
keep-notes/lib/ai/services/embedding.service.ts
Normal file
@@ -0,0 +1,224 @@
|
||||
/**
|
||||
* Embedding Service
|
||||
* Generates vector embeddings for semantic search and similarity analysis
|
||||
* Uses text-embedding-3-small model via OpenAI (or Ollama alternatives)
|
||||
*/
|
||||
|
||||
import { getAIProvider } from '../factory'
|
||||
|
||||
export interface EmbeddingResult {
|
||||
embedding: number[]
|
||||
model: string
|
||||
dimension: number
|
||||
}
|
||||
|
||||
/**
|
||||
* Service for generating and managing text embeddings
|
||||
*/
|
||||
export class EmbeddingService {
|
||||
private readonly EMBEDDING_MODEL = 'text-embedding-3-small'
|
||||
private readonly EMBEDDING_DIMENSION = 1536 // OpenAI's embedding dimension
|
||||
|
||||
/**
|
||||
* Generate embedding for a single text
|
||||
*/
|
||||
async generateEmbedding(text: string): Promise<EmbeddingResult> {
|
||||
if (!text || text.trim().length === 0) {
|
||||
throw new Error('Cannot generate embedding for empty text')
|
||||
}
|
||||
|
||||
try {
|
||||
const provider = getAIProvider()
|
||||
|
||||
// Use the existing getEmbeddings method from AIProvider
|
||||
const embedding = await provider.getEmbeddings(text)
|
||||
|
||||
// Validate embedding dimension
|
||||
if (embedding.length !== this.EMBEDDING_DIMENSION) {
|
||||
}
|
||||
|
||||
return {
|
||||
embedding,
|
||||
model: this.EMBEDDING_MODEL,
|
||||
dimension: embedding.length
|
||||
}
|
||||
} catch (error) {
|
||||
console.error('Error generating embedding:', error)
|
||||
throw new Error(`Failed to generate embedding: ${error}`)
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Generate embeddings for multiple texts in batch
|
||||
* More efficient than calling generateEmbedding multiple times
|
||||
*/
|
||||
async generateBatchEmbeddings(texts: string[]): Promise<EmbeddingResult[]> {
|
||||
if (!texts || texts.length === 0) {
|
||||
return []
|
||||
}
|
||||
|
||||
// Filter out empty texts
|
||||
const validTexts = texts.filter(t => t && t.trim().length > 0)
|
||||
|
||||
if (validTexts.length === 0) {
|
||||
return []
|
||||
}
|
||||
|
||||
try {
|
||||
const provider = getAIProvider()
|
||||
|
||||
// Batch embedding using the existing getEmbeddings method
|
||||
const embeddings = await Promise.all(
|
||||
validTexts.map(text => provider.getEmbeddings(text))
|
||||
)
|
||||
|
||||
return embeddings.map(embedding => ({
|
||||
embedding,
|
||||
model: this.EMBEDDING_MODEL,
|
||||
dimension: embedding.length
|
||||
}))
|
||||
} catch (error) {
|
||||
console.error('Error generating batch embeddings:', error)
|
||||
throw error
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Calculate cosine similarity between two embeddings
|
||||
* Returns value between -1 and 1, where 1 is identical
|
||||
*/
|
||||
calculateCosineSimilarity(embedding1: number[], embedding2: number[]): number {
|
||||
if (embedding1.length !== embedding2.length) {
|
||||
throw new Error('Embeddings must have the same dimension')
|
||||
}
|
||||
|
||||
let dotProduct = 0
|
||||
let magnitude1 = 0
|
||||
let magnitude2 = 0
|
||||
|
||||
for (let i = 0; i < embedding1.length; i++) {
|
||||
dotProduct += embedding1[i] * embedding2[i]
|
||||
magnitude1 += embedding1[i] * embedding1[i]
|
||||
magnitude2 += embedding2[i] * embedding2[i]
|
||||
}
|
||||
|
||||
magnitude1 = Math.sqrt(magnitude1)
|
||||
magnitude2 = Math.sqrt(magnitude2)
|
||||
|
||||
if (magnitude1 === 0 || magnitude2 === 0) {
|
||||
return 0
|
||||
}
|
||||
|
||||
return dotProduct / (magnitude1 * magnitude2)
|
||||
}
|
||||
|
||||
/**
|
||||
* Calculate similarity between an embedding and multiple other embeddings
|
||||
* Returns array of similarities
|
||||
*/
|
||||
calculateSimilarities(
|
||||
queryEmbedding: number[],
|
||||
targetEmbeddings: number[][]
|
||||
): number[] {
|
||||
return targetEmbeddings.map(embedding =>
|
||||
this.calculateCosineSimilarity(queryEmbedding, embedding)
|
||||
)
|
||||
}
|
||||
|
||||
/**
|
||||
* Find most similar embeddings to a query
|
||||
* Returns top-k results with their similarities
|
||||
*/
|
||||
findMostSimilar(
|
||||
queryEmbedding: number[],
|
||||
targetEmbeddings: Array<{ id: string; embedding: number[] }>,
|
||||
topK: number = 10
|
||||
): Array<{ id: string; similarity: number }> {
|
||||
const similarities = targetEmbeddings.map(({ id, embedding }) => ({
|
||||
id,
|
||||
similarity: this.calculateCosineSimilarity(queryEmbedding, embedding)
|
||||
}))
|
||||
|
||||
// Sort by similarity descending and return top-k
|
||||
return similarities
|
||||
.sort((a, b) => b.similarity - a.similarity)
|
||||
.slice(0, topK)
|
||||
}
|
||||
|
||||
/**
|
||||
* Get average embedding from multiple embeddings
|
||||
* Useful for clustering or centroid calculation
|
||||
*/
|
||||
averageEmbeddings(embeddings: number[][]): number[] {
|
||||
if (embeddings.length === 0) {
|
||||
throw new Error('Cannot average empty embeddings array')
|
||||
}
|
||||
|
||||
const dimension = embeddings[0].length
|
||||
const average = new Array(dimension).fill(0)
|
||||
|
||||
for (const embedding of embeddings) {
|
||||
if (embedding.length !== dimension) {
|
||||
throw new Error('All embeddings must have the same dimension')
|
||||
}
|
||||
|
||||
for (let i = 0; i < dimension; i++) {
|
||||
average[i] += embedding[i]
|
||||
}
|
||||
}
|
||||
|
||||
// Divide by number of embeddings
|
||||
return average.map(val => val / embeddings.length)
|
||||
}
|
||||
|
||||
/**
|
||||
* Serialize embedding to JSON-safe format (for storage)
|
||||
*/
|
||||
serialize(embedding: number[]): string {
|
||||
return JSON.stringify(embedding)
|
||||
}
|
||||
|
||||
/**
|
||||
* Deserialize embedding from JSON string
|
||||
*/
|
||||
deserialize(jsonString: string): number[] {
|
||||
try {
|
||||
const parsed = JSON.parse(jsonString)
|
||||
if (!Array.isArray(parsed)) {
|
||||
throw new Error('Invalid embedding format')
|
||||
}
|
||||
return parsed
|
||||
} catch (error) {
|
||||
console.error('Error deserializing embedding:', error)
|
||||
throw new Error('Failed to deserialize embedding')
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Check if a note needs embedding regeneration
|
||||
* (e.g., if content has changed significantly)
|
||||
*/
|
||||
shouldRegenerateEmbedding(
|
||||
noteContent: string,
|
||||
lastEmbeddingContent: string | null,
|
||||
lastAnalysis: Date | null
|
||||
): boolean {
|
||||
// If no previous embedding, generate one
|
||||
if (!lastEmbeddingContent || !lastAnalysis) {
|
||||
return true
|
||||
}
|
||||
|
||||
// If content has changed more than 20% (simple heuristic)
|
||||
const contentChanged =
|
||||
Math.abs(noteContent.length - lastEmbeddingContent.length) / lastEmbeddingContent.length > 0.2
|
||||
|
||||
// If last analysis is more than 7 days old
|
||||
const daysSinceAnalysis = (Date.now() - lastAnalysis.getTime()) / (1000 * 60 * 60 * 24)
|
||||
const isStale = daysSinceAnalysis > 7
|
||||
|
||||
return contentChanged || isStale
|
||||
}
|
||||
}
|
||||
|
||||
// Singleton instance
|
||||
export const embeddingService = new EmbeddingService()
|
||||
Reference in New Issue
Block a user