AI-Assisted Compliance Dossier Generator (SaaS)

Overview

A multi-tenant SaaS that assembles regulatory traceability dossiers from heterogeneous source documents (invoices, transport notes, intake/receipt records, official minutes and permits). It combines a modern React front end, a serverless Convex backend, and AI-assisted extraction to turn uploaded PDFs into structured, exportable compliance files.

The Challenge

Compiling a complete traceability dossier by hand means re-keying data from dozens of scanned invoices, transport documents and approval records into a consistent structure, slow, error-prone, and hard to audit. The product automates collection, extraction and assembly while keeping each organisation’s data isolated.

What We Built

A React 19 + Vite single-page app with a shadcn/ui (Radix) component system, organised into pages, providers and a typed router. The Convex backend exposes domain modules for the real document types in play, dosars, documents, facturi (invoices), transports/transportWasteItems, receptii (receipts), proceseVerbale (official minutes), avizuri (permits), depozit, ficds and uploads, plus emailIngestion for inbound document capture, orgSettings for per-tenant configuration, and an ai.ts module wiring OpenAI-based extraction. Auth and organisation tenancy run on Clerk; PDFs are parsed and previewed client-side with pdfjs-dist, react-pdf and unpdf, and bundled for export with jszip. A separate email-worker handles ingestion outside the main app.

Technologies & Approach

React 19/TypeScript/Vite for the UI; Convex for a reactive, serverless data and function layer; Clerk for authentication and multi-org tenancy; OpenAI for extracting structured fields from uploaded documents; a PDF toolchain (pdfjs-dist/react-pdf/unpdf) for rendering and parsing; Zod for schema validation. The architecture keeps document logic in clearly-named backend modules so each compliance artefact maps to its own function set.

Outcome / Impact

An actively developed product that demonstrates an end-to-end document-automation stack: ingest (upload + email), AI extraction, structured storage, and dossier export, all multi-tenant from day one.

Capabilities Demonstrated

• End-to-end document-automation SaaS: ingest, extract, assemble, export
• AI-assisted structured data extraction from PDFs
• Multi-tenant architecture with Clerk org-based isolation
• Serverless reactive backend design with Convex
• Modern React 19 + Tailwind v4 + shadcn/ui front end