🛡 Compliance AI Product Validator

Overview

This project was developed for a client to simplify and automate the complex process of product compliance verification in Europe. Traditionally, compliance checks required 1-2 weeks of manual work by specialists, costing up to €1000 per product.

With this system, the entire process can be completed in less than 1 minute, fully automated and repeatable.

---

What it does

• Upload a certification PDF and product images
• The system scans the document for required tests and keywords
• Extracts and validates images inside the PDF
• Compares uploaded product photos with official certification images
• Produces a clear report that confirms whether the product passes compliance checks
• Reports can be accessed from a dashboard or sent by email automatically

---

My role

I was responsible for the entire system design and implementation:

• Designing the backend services and infrastructure
• Integrating multiple AI models for document and image understanding
• Optimizing the workflow to guarantee results in under one minute
• Building the dashboard for clients to track past analyses and reports
• Setting up AWS cloud infrastructure, monitoring, and automated scaling
• Generating professional PDF reports automatically for client delivery

---

Why it matters

• Reduced compliance verification from 2 weeks → under 1 minute
• Cut costs by over 90% compared to manual checks
• Enabled the client to sell compliance checks as a SaaS product
• Improved customer experience with instant reports and dashboards
• Scalable to handle dozens of products in parallel without extra staff

---

Tech Stack

• Backend: Python, FastAPI
• Cloud: AWS (Lambda, S3, ECS, EventBridge, SES)
• AI/LLMs: OpenAI (Vision), Google Gemini, Mistral
• Agent Orchestration: LangGraph, custom pipelines
• Data Processing: PDF parsing, image extraction and comparison
• Reporting: Puppeteer on AWS Lambda for PDF report generation
• Observability: Logging, benchmarking, monitoring for audit requirements

---

Status

This project was built for a private client and is not open-sourced. The repository contains a case study and technical overview of the solution.