POC Sophia

POC Sophia is an artificial intelligence tool designed for statutory auditors as part of the HackAudit hackathon. The application automates documentary pre-analysis of accounting documents (FEC, invoices, delivery notes) via N8N workflows with LLM analysis, while maintaining mandatory human supervision visible on every screen. Compliant with NEP 230 standard, the tool generates 4 deliverables: completeness checklist, documentary signals table, pre-analysis summary, and traceability journal (immutable audit trail). Three audit scenarios are covered: revenue sampling, year-end cutoff, and financial statements reconciliation.

Challenges

• Ensuring NEP 230 compliance with mandatory human supervision on every screen
• Asynchronous AI analysis orchestration via N8N with status polling
• Full explainability: each signal includes reasoning, source document, and triggered rule
• 48-hour development for the HackAudit hackathon

Solutions

• N8N architecture as temporary processing service with results imported to Supabase
• Signal rules catalog (DOC-*, SCA-*, CUT-*, PTG-*) with complete traceability
• Server-side PDF/XLSX regeneration after auditor decision validation
• Supabase RLS for data isolation per license with 48-hour TTL on N8N results

Results

• 4 functional deliverables: Checklist, Signals, Summary, Traceability Journal
• 3 complete audit scenarios with fictional data (ACME Distribution SAS)
• NEP 230 compliance and integrated explainability framework
• HackAudit hackathon finalist

Technologies

Next.js 14 · TypeScript · Supabase · N8N · Radix UI · Tailwind CSS · pdfmake