Compliance AI Product Validator

The problem

A European compliance specialist firm was charging clients €1,000 per product to manually verify certification documents. Specialists spent 1–2 weeks per case cross-checking PDFs, validating required tests, comparing product photos against official certification imagery, and writing the final report. The process was slow, expensive, and impossible to scale.

The approach

End-to-end backend system that ingests a certification PDF + product images and produces a verdict in under a minute.

• Multi-model AI pipeline. OpenAI Vision, Gemini, and Mistral each handle the document type they’re best at — text extraction, structural validation, image comparison — coordinated by LangGraph so each step has the right tool and clean fallbacks.
• AWS-native infra. Lambda for async processing, ECS for heavy jobs, S3 for artifacts, EventBridge for orchestration, SES for client delivery. Scales horizontally to dozens of products in parallel without staffing changes.
• PDF report generation. Puppeteer on Lambda renders the final compliance report as a branded PDF that lands in the client’s inbox automatically.
• Client dashboard. Past analyses, status tracking, downloadable reports — productized so the firm can sell it.

The outcome

• Verification time: 2 weeks → under 1 minute
• Cost per check: >90% reduction vs. manual specialist work
• Enabled the firm to launch compliance checks as a SaaS product — new revenue line, not just internal efficiency
• Scales to dozens of products in parallel without additional staff

My role

End-to-end system design and implementation — backend services, multi-model AI orchestration, AWS infrastructure, monitoring, dashboard, PDF report generation. Built for a private client; the public repo is the technical case study.