DocAnalyzer Pro (Enterprise Edition) 🚀

High-Performance Document Intelligence Platform by Xentro AI

DocAnalyzer Pro is a hybrid AI platform designed to automate complex financial document processing. Unlike standard wrappers, it uses a custom-compiled C++ Vision Engine to pre-process low-quality images (deskewing, binarization, upscaling) before routing data to a Retrieval-Augmented Generation (RAG) pipeline for semantic analysis.

---

⚡ Deep Core Technology

1. 🧠 The C++ Vision Engine (docproc)

We do not rely on Python for heavy image processing. The core engine is written in C++17 using OpenCV and Tesseract 5.

• "Super-Resolution" Pipeline: Automatically upscales low-res receipt photos (36KB) by 300% using bicubic interpolation.
• Heuristic Rotation Logic: Solves the "rotated text" problem by testing 4 orientation vectors (0°, 90°, 180°, 270°) and selecting the optimal angle based on OCR confidence scores.
• Adaptive Thresholding: Uses local block analysis to remove shadows and wood-grain backgrounds from camera photos.

2. 🛡️ Enterprise Architecture

The system follows a strict Vertical Slice Architecture:

• Orchestrator: Python subprocess management with timeout handling and memory leak prevention.
• Dual-Database System:
  • SQL (SQLite): Transactional records, file hashing (MD5), and batch history.
  • Vector Store (ChromaDB): Semantic embeddings for the "Chat with Data" feature.
• Context Isolation: Strict RAG filtering ensures the AI answers questions only about the active document, preventing data leakage between clients.

3. 🎨 "Neural HUD" Interface

A custom-built reactive UI on top of Streamlit:

• Dynamic Theme Engine: Hot-swaps CSS variables for Neon Dark Mode / Corporate Light Mode.
• Visual Debugging: Exposes the raw C++ computer vision output ("CEO Vision") to build user trust.
• Batch Telemetry: Aggregates financial totals across hundreds of documents in real-time.

---

🏗️ Architecture Diagram

graph TD
    A[User Upload] -->|PDF/JPG/CSV| B(Python Orchestrator)
    B -->|MD5 Hash| C{Cache Hit?}
    C -- Yes --> D[Load from SQL DB]
    C -- No --> E{File Type?}
    E -- Image/PDF --> F[C++ Vision Engine]
    F -->|OpenCV Pre-process| G[Tesseract OCR]
    E -- CSV --> H[Pandas Parser]
    G --> I[Gemini 1.5 Flash]
    H --> I
    I --> J[Structured JSON]
    J --> K[(SQLite + ChromaDB)]
    K --> L[Streamlit Dashboard]

Loading

---

🚀 Features at a Glance

Feature	Capability
Batch Processing	Drag & drop 50+ files; auto-loops and aggregates totals.
Smart Renaming	Renames scan_001.jpg → 2025-11-24_HotelNewYork_Receipt.jpg.
Financial Trends	Interactive Altair charts showing spending by Vendor vs. Date.
Confidence Heatmap	Color-coded table rows (Green/Red) based on AI certainty.
Auto-Correction	Built-in Data Editor allows humans to fix OCR errors before export.

Core Processing (v2)

Feature	Capability
Hybrid C++ Engine	Custom OpenCV pipeline for deskewing & upscaling receipt photos.
Batch Processing	Parallel processing of 50+ mixed documents (PDF/JPG/CSV).
Smart Renaming	Auto-renames files: scan_001.jpg → 2025-11-24_Vendor_Invoice.jpg.

Enterprise Intelligence (v3) 🆕

Feature	Capability
🛡️ Anomaly Auditor	Fraud Detection: Flags invoices with >40% price deviation or unknown vendors against historical data.
🌍 Global Brain	Cross-Doc Intelligence: Ask questions across your entire database (e.g., "Total spend on Microsoft last year?").
🔒 Privacy Vault	GDPR Compliance: Auto-redacts PII (Names, Phones) from JSON exports for safe sharing.
📊 Financial Analytics	Interactive charts showing spending trends by Vendor and Time.

---

🛠️ Installation & Setup

Prerequisites

• Linux/Mac (or WSL2 on Windows)
• C++ Compiler (g++ or clang)
• Python 3.10+
• Tesseract & OpenCV libraries

1. Clone & Build

git clone https://github.com/xentro-ai/docanalyzer-pro.git
cd docanalyzer-pro

# Build the C++ Engine
mkdir build && cd build
cmake .. && make
cd ..

2. Install Dependencies

pip install -r requirements.txt

3. Run the Platform

streamlit run src/python/app.py

---

🔮 Future Roadmap (v3.0)

• Docker Containerization: Deploy as a microservice container.
• API Endpoint: Expose FastAPI routes for external integrations.
• Local LLM Fallback: Support for DeepSeek-Coder when offline.

---

Built with ❤️ by Xentro AI Hybrid Automation Experts