What I design as an AI architect
An enterprise AI solution is much more than wiring a model to an API. Which data reaches the model, how the output is validated, which systems it writes to and how all of it is monitored are architectural decisions. As an AI architect and project specialist I make those decisions together with the business goal.
- Use-case selection: frequent, rule-bound tasks where errors are measurable
- Data and context: RAG (answers grounded in internal documents), document parsing, structured output
- Integration: controlled access to SAP via BAPI/OData and to other systems via REST APIs
- Security and compliance: data protection (KVKK/GDPR), masking, authorisation, logging and audit trail
- Measurement: accuracy, processing time, human-intervention rate and cost metrics
ERP and AI: realistic use cases
- Extracting data from supplier invoices and order documents, with approved transfer to SAP
- Classifying customer emails and turning them into order or request records
- Summarising SAP reports in natural language and explaining variances
- Question-answering assistants over internal process documentation
From pilot to production
Every project starts with a narrow, measurable pilot. I define the success criteria (for example accuracy and time saved) up front and widen the scope only once they are met. I also build a provider-independent layer so the system keeps working even when the model changes.
Architecture backed by software expertise
I do not leave the architecture on paper; with my software expertise in PHP/Laravel and REST APIs I build the integration itself. The AI assistant on this site follows the same approach: keys and rules stay on the server, and requests are rate-limited.
