Multimodal eDiscovery for Litigation Support

import { Mixpeek } from 'mixpeek';

const client = new Mixpeek({ apiKey: process.env.MIXPEEK_API_KEY });

// Process matter discovery
await client.buckets.connect({
  collection_id: 'matter-2024-0892',
  bucket_uri: 's3://ediscovery/matter-2024-0892/',
  extractors: [
    'document-parser',      // PDFs, Word, Excel
    'email-parser',         // PST, MSG, EML
    'speech-to-text',       // Audio/video recordings
    'image-ocr',            // Screenshots, scanned docs
    'video-frame-analysis'  // Visual content in videos
  ],
  settings: {
    preserve_metadata: true,   // Retain original metadata
    generate_hash: true,       // Deduplication
    privilege_markers: ['PRIVILEGED', 'ATTORNEY-CLIENT']
  }
});

// Find potentially privileged content
const privileged = await client.retrieve({
  collection_id: 'matter-2024-0892',
  query: {
    type: 'text',
    text: 'legal advice attorney consultation litigation strategy'
  },
  filters: {
    content_type: { $in: ['email', 'document'] }
  },
  return_fields: ['content', 'participants', 'date', 'source_file'],
  limit: 1000
});

// Route for attorney privilege review
const forReview = privileged.filter(
  doc => doc.privilege_score > 0.7
);

// Search for evidence related to specific issues
async function searchMatter(matterId: string, issue: string) {
  const results = await client.retrieve({
    collection_id: `matter-${matterId}`,
    query: {
      type: 'text',
      text: issue // e.g., "discussions about pricing changes Q3 2023"
    },
    return_fields: [
      'content', 'source_file', 'date',
      'participants', 'timestamp', 'relevance_score'
    ],
    limit: 500
  });

  return {
    documents: results.filter(r => r.type === 'document'),
    emails: results.filter(r => r.type === 'email'),
    recordings: results.filter(r => r.type === 'audio' || r.type === 'video'),
    total_hits: results.length
  };
}

// Generate audit trail for defensibility
const auditReport = {
  matter_id: matterId,
  search_methodology: 'Semantic similarity + keyword boolean',
  total_documents: corpus.total,
  documents_reviewed: corpus.reviewed,
  ai_assisted_review: {
    model: 'mixpeek-legal-v2',
    confidence_threshold: 0.85,
    human_review_rate: 0.15 // 15% sampled for QC
  },
  privilege_log_generated: true,
  date_certified: new Date().toISOString()
};

await generateDefensibilityReport(auditReport);