Back

Deviscan

AI-powered quote extraction Micro-SaaS

Next.js 16FastAPIGeminiClaudeFactur-XSupabaseCloudflare R2LemonSqueezy

Context

SMBs receive dozens of supplier quotes every month — as PDFs, sometimes scanned, never in the same format. Buyers spend hours manually copying line items into Excel to compare offers.

The problem

No affordable solution to automatically extract structured data from a quote PDF (lines, quantities, unit prices, VAT, conditions) and make it actionable.

Technical solution

Deviscan is a micro-SaaS that turns a quote PDF into structured data in seconds.

Architecture

  • Frontend: Next.js 16 (App Router), TypeScript, Tailwind CSS, Zustand
  • Backend: FastAPI (Python), SQLAlchemy 2 async, Alembic
  • Auth: Supabase (JWT via JWKS)
  • AI: Model cascade — Gemini Flash-Lite → Gemini Pro → Claude (fallback)
  • Storage: Cloudflare R2 (S3-compatible)
  • Billing: LemonSqueezy (HMAC webhooks)

Key features

  • Multi-file upload (PDF, DOCX, images)
  • AI extraction with scanned PDF handling (image preprocessing)
  • Manual review and edit of extracted line items
  • Export to Excel, CSV and Factur-X (EN 16931 standard)
  • Side-by-side quote comparison
  • Plan-based quotas (free / pro / business)
  • Batch multi-document export

Notable technical details

  • AI cascade with automatic provider fallback
  • CII XML generation for Factur-X / PDF-A compliance
  • Per-endpoint rate limiting (slowapi)
  • Backend test suite covering auth, billing, extraction, export