AI-Driven Intelligent Automation Engine for Multicountry Export Certification

What began as a targeted request to automate export certificate generation grew into a broader transformation: an intelligent document automation platform built on top of the CS IT-enterprise framework. The initial landscape was fragmented, with disjointed workflows, manual data entry, and unstructured files (including PDF invoices, packing lists, and transport tables) circulating across departments. Delays and errors were frequent and costly.

We transformed this process by layering intelligent automation on top of the CS IT-enterprise framework. Devox engineered a modular web platform with embedded OCR, AI-driven data extraction, and configurable certificate generation logic. Built with FastAPI and deployed on cloud-native infrastructure, the system now manages structured document flows tailored to 34 destination countries.

The platform allows business users to manage export documentation through a unified interface. At the same time, the intelligent backend handles complexity, including parsing, field validation, multi-language support, template matching, and error reporting. All logic is driven by flexible JSON configurations, allowing rapid updates as regulations evolve.

• Solution Architect (document workflow domain)
• 2 Full-Stack Engineers (React, Node.js)
• QA Automation Engineer
• DevOps Engineer (Docker, CI/CD, monitoring)
• Business Analyst (export logistics & compliance)
• Project Manager

• Fragmented data inputs. Certificates relied on three separate document types — invoices (PDF), packing lists (PDF), and transport tables (XLSX/XML), each with its own structure and inconsistent formatting.
• Manual data extraction. All key fields (e.g., sender address, product details, seal numbers) were manually retrieved from unstructured files, which increased error rates and processing time.
• Country-specific variability. Each of the 34 supported countries required a unique combination of document templates (DOCX, XLSX), fields, and formatting, resulting in over 85 distinct variants in total.
• No structured validation flow. There was no mechanism to pre-validate input files or fields, which meant errors were only discovered during or after certificate generation.
• Rigid scaling model. Support for new countries or templates required direct changes to the system logic, as there was no dynamic configuration engine in place.
• No feedback on OCR confidence. The system initially lacked mechanisms to flag fields extracted with low certainty, leaving users unaware of potential inaccuracies.
• Disconnected certification logic. Generated certificates were not automatically linked to the original input documents or configurations, which complicated traceability and auditing.

• Frontend: React, i18next (multi-language support), React Router, Material UI / Ant Design
• Backend: FastAPI, Python, Pydantic
• AI & OCR: Google Gemini API, pytesseract, pandas, openpyxl
• Document Generation: docxtpl (for DOCX), openpyxl (for XLSX)
• Configuration Management: JSON-based templates per country
• Database: PostgreSQL (hosted in the cloud)
• DevOps & Deployment: Docker, docker-compose, AWS ECS / Fargate / Cloud Run
• Storage: AWS S3 / Azure Blob / Google Cloud Storage
• Authentication & Security: JWT token-based auth, secrets management via cloud services
• Monitoring & Logging: Cloud-native observability (CloudWatch / Stackdriver / etc.)

Devox Software designed and built a modular certificate generation platform on top of CS It-enterprise, combining traditional document workflow logic with intelligent automation.

• Dynamic document intake. Users upload an invoice (PDF), a packing list (PDF), and a transport table (XLSX or XML). The system extracts relevant data using OCR (pytesseract, Gemini API) and table parsers (pandas, openpyxl).
• Country-specific configuration engine. Each of the 34 supported countries is mapped to a JSON configuration file, which defines the required fields, validation rules, and output templates. This enables the addition of new destinations without changing the core logic.
• AI-powered field extraction. The system uses confidence scoring to validate OCR output and flags fields with low certainty for manual review.
• Certificate rendering pipeline. Based on validated input, the system generates all necessary documents (DOCX/XLSX), including import certificates, manufacturer declarations, transit documents, “No wood” certificates, and annexes, all of which are country-compliant.
• Web-based UI. A desktop-oriented React interface enables authenticated users to manage document uploads, input dynamic form fields, and download generated files, with full support for two languages (UA and the target country).
• Cloud-native deployment. The solution is containerized (using Docker), API-first (built with FastAPI), and integrates with cloud storage and database services for scalability, traceability, and secure access.

BUSINESS OUTCOMES

• 80% reduction in manual effort. Certificate preparation time dropped significantly due to automated field extraction and template rendering.
• Faster document readiness. Certificates are now generated in minutes, rather than hours, accelerating export readiness.
• Higher accuracy. Confidence scoring and structured validation minimized human error and improved regulatory compliance.
• Scalable to 34+ countries. New certificate types or destinations can be supported without development overhead.

TECHNICAL OUTCOMES

• Fully modular architecture. Web UI, OCR engine, template renderer, and configuration layer operate independently, enabling parallel updates.
• Dynamic config management. All logic (fields, languages, templates) is driven by country-level JSON files, allowing fast adaptation to regulatory changes.
• Integrated validation and traceability. Every output file can be traced back to its input documents, with clear error logs and status feedback.
• Cloud-ready deployment. Dockerized microservices operate in isolated environments with cloud-based storage and a PostgreSQL database, supporting horizontal scaling and secure data flow.