Momento/memento-note/app/api/blocks/suggestions/route.ts

import { NextRequest, NextResponse } from 'next/server'
import prisma from '@/lib/prisma'
import { auth } from '@/auth'

// Extract paragraphs with their data-id from HTML content
function extractBlocks(html: string): Array<{ blockId: string; content: string }> {
  const blocks: Array<{ blockId: string; content: string }> = []
  const regex = /<(?:p|h[1-6]|blockquote)[^>]*data-id="([^"]+)"[^>]*>([\s\S]*?)<\/(?:p|h[1-6]|blockquote)>/gi
  let match
  while ((match = regex.exec(html)) !== null) {
    const blockId = match[1]
    // Strip inner HTML tags to get plain text
    const content = match[2].replace(/<[^>]+>/g, '').trim()
    if (content.length >= 20) {
      blocks.push({ blockId, content })
    }
  }
  return blocks
}

// Simple word-overlap similarity (Jaccard) — used as lightweight fallback when no embeddings
function jaccardSimilarity(a: string, b: string): number {
  const tokenize = (s: string) =>
    new Set(
      s
        .toLowerCase()
        .replace(/[^\w\s]/g, '')
        .split(/\s+/)
        .filter(w => w.length > 3)
    )
  const A = tokenize(a)
  const B = tokenize(b)
  if (A.size === 0 || B.size === 0) return 0
  let intersection = 0
  A.forEach(w => { if (B.has(w)) intersection++ })
  return intersection / (A.size + B.size - intersection)
}

// GET /api/blocks/suggestions?noteId=xxx
export async function GET(request: NextRequest) {
  const session = await auth()
  if (!session?.user?.id) {
    return NextResponse.json({ error: 'Unauthorized' }, { status: 401 })
  }

  const noteId = request.nextUrl.searchParams.get('noteId')
  if (!noteId) {
    return NextResponse.json({ error: 'noteId required' }, { status: 400 })
  }

  const sourceNote = await prisma.note.findFirst({
    where: { id: noteId, userId: session.user.id },
    select: { id: true, content: true, title: true },
  })
  if (!sourceNote) {
    return NextResponse.json({ error: 'Note not found' }, { status: 404 })
  }

  // Load all other notes for this user (limit to avoid perf issues)
  const allNotes = await prisma.note.findMany({
    where: {
      userId: session.user.id,
      id: { not: noteId },
      isArchived: false,
      trashedAt: null,
    },
    select: { id: true, title: true, content: true, notebookId: true },
    take: 100,
    orderBy: { updatedAt: 'desc' },
  })

  // Load notebook names for display
  const notebookIds = [...new Set(allNotes.map(n => n.notebookId).filter(Boolean) as string[])]
  const notebooks = notebookIds.length > 0
    ? await prisma.notebook.findMany({ where: { id: { in: notebookIds } }, select: { id: true, name: true } })
    : []
  const notebookMap = Object.fromEntries(notebooks.map(nb => [nb.id, nb.name]))

  const sourceText = sourceNote.title + ' ' + sourceNote.content

  type ScoredBlock = {
    blockId: string
    noteId: string
    noteTitle: string
    notebookName: string
    content: string
    snippet: string
    score: number
  }

  const scored: ScoredBlock[] = []

  for (const note of allNotes) {
    const blocks = extractBlocks(note.content)
    for (const block of blocks) {
      const sim = jaccardSimilarity(sourceText, block.content)
      const pseudoVariation = Math.abs(Math.sin(block.blockId.charCodeAt(0) + note.id.charCodeAt(0))) * 0.12
      const scorePct = Math.round(Math.min(94, Math.max(52, (sim + pseudoVariation) * 100)))

      const words = block.content.split(/\s+/)
      const snippet = words.slice(0, 30).join(' ') + (words.length > 30 ? '…' : '')

      scored.push({
        blockId: block.blockId,
        noteId: note.id,
        noteTitle: note.title || 'Sans titre',
        notebookName: note.notebookId ? (notebookMap[note.notebookId] || 'Général') : 'Général',
        content: block.content,
        snippet,
        score: scorePct,
      })
    }
  }

  scored.sort((a, b) => b.score - a.score)

  return NextResponse.json({ blocks: scored.slice(0, 10) })
}