natgluons / AI-docs-analyzer-API

Automate invoice analysis and identity verification, built with an open-source multimodal LLM and OCR (DocTR/TrOCR), using FastAPI, Supabase, PgVector, and Neo4j.

AI Document Intelligence API: Invoice Analyzer & ID Verifier

About

The AI Document Intelligence API is a smart document processing system that combines invoice analysis and identity document verification in a unified backend. This project is built to address practical problems in financial operations, document automation, and fraud detection by leveraging OCR, vector search, and knowledge graphs.

Manual document validation is time-consuming and error-prone. This API offers automated pipelines for extracting structured data, detecting duplicate or suspicious documents, and matching key fields to existing records or knowledge graphs, making it useful for fintech, procurement, and KYC applications.

---

Tech Stack

Layer	Tech Used	Purpose
Backend API	FastAPI	API server and document routing
Database	Supabase (PostgreSQL)	Structured data + file storage
Embedding	PgVector	Vector search and semantic similarity
OCR	DocTR / TrOCR	Multimodal text extraction from images/PDFs
Graph	Neo4j (optional)	Fraud relationship analysis and vendor graphs
Cloud	Microsoft Azure	OCR inference hosting, file uploads

---

Architecture Diagram

graph TD
    A[User Uploads Document] --> B[FastAPI Upload Endpoint]
    B --> C[OCR Module - DocTR or TrOCR]
    C --> D[Extracted Text]
    D --> E[Field Extraction and Classification]
    E --> F[Vector Embedding with PgVector]
    F --> G[Similarity Search and Duplicate Detection]
    E --> H[Save Fields to Supabase DB]
    B --> I[Upload Original File to Supabase Storage]
    G --> J[Optional Fraud and Vendor Graph - Neo4j]

Loading

---

Database Schema (Supabase)

documents

Column	Type	Description
id	UUID	Primary key
user_id	UUID	Reference to user
doc_type	TEXT	'invoice', 'id_card', 'certificate'
file_url	TEXT	Supabase storage URL
upload_date	TIMESTAMP	Document upload date
status	TEXT	'processed', 'error', etc.

extracted_fields

Column	Type	Description
id	UUID	Primary key
doc_id	UUID	Foreign key to documents
field_name	TEXT	e.g., 'vendor_name', 'amount'
field_value	TEXT	Extracted value
confidence	FLOAT	Confidence score

embeddings

Column	Type	Description
id	UUID	Primary key
doc_id	UUID	Foreign key to documents
vector	VECTOR	PgVector-embedded representation
type	TEXT	'invoice_meta', 'id_identity'

---

Key Features

• Multimodal OCR: Handles both printed and handwritten text in various document formats.
• Auto Classification: Determines if input is invoice, ID card, or certificate.
• PgVector Search: Detects similar documents (e.g. fake IDs, duplicate invoices).
• Vendor Knowledge Graph (Optional): Build a graph of vendors and transaction patterns.
• Modular Architecture: Easily plug in new document types or rules.

---

Example Use Cases

• Fintech platforms automating invoice uploads & verification
• KYC onboarding systems validating national ID or license
• Procurement automation tools preventing invoice fraud
• Academic verification for certificates and diplomas

---

Coming Soon

• Graph dashboard with Neo4j or Graphistry
• React frontend for upload and results
• PDF highlighting of extracted fields