# Muhammed Durakovic — Solutions Architect for Intelligent Systems

> I design AI, software, data, and infrastructure as one working system. When AI has to work in the real world, talk to me.

## About

Muhammed Durakovic is a Solutions Architect and AI Engineer based in Graz, Austria. He holds a Dipl.-Ing. in Software Engineering & Management, specializing in Intelligent Systems, from Graz University of Technology (TU Graz).

He understands critical enterprise systems quickly, then designs AI transformations that fit their data, constraints, infrastructure, security, and operations. He owns the path from technical discovery and executive-level architecture proposals through implementation, deployment, and adoption. He works across models and vendors, including cloud and self-hosted systems.

## Profile sections

- [Homepage and positioning](https://durakovic.ai/)
- [Open-source work](https://durakovic.ai/#open-source)
- [Recent roles](https://durakovic.ai/#experience)
- [AI practice](https://durakovic.ai/#ai)
- [Contact](https://durakovic.ai/#contact)

## Open-source work

- [oraclemcp](https://github.com/MuhDur/oraclemcp): safe-by-default Oracle access for AI agents. Read-only by default, with a fail-closed SQL guard, OAuth 2.1, and an audit trail. Listed in the official MCP registry and published on crates.io.
- [rust-oracledb](https://github.com/MuhDur/rust-oracledb): a clean-room, pure-Rust async Oracle driver with 2,462 conformance tests passing. Published on crates.io as `oracledb`.
- [plsql-intelligence](https://github.com/MuhDur/plsql-intelligence): package-aware Oracle PL/SQL code intelligence for parsing, dependency analysis, lineage, SAST, and change-impact analysis.
- [invoicekit](https://github.com/MuhDur/invoicekit): an Apache-2.0 engine for global B2B e-invoicing formats, including Factur-X, ZUGFeRD, XRechnung, UBL, Peppol BIS, FatturaPA, KSeF, ZATCA, and CFDI.

## AI practice

Regular tools include Claude, Codex, Gemini, Nous Research Hermes, and Qwen. The working stack includes local models where privacy, control, latency, or cost makes local inference the better fit. Other systems evaluated in practice include Kimi, MiniMax, GLM, Grok, and Cursor.

## Contact and profiles

- Email: [hello@durakovic.ai](mailto:hello@durakovic.ai)
- LinkedIn: [Muhammed Durakovic](https://www.linkedin.com/in/muhammed-durakovic-70047516a)
- GitHub: [MuhDur](https://github.com/MuhDur)
- X: [@itsmuhdur](https://x.com/itsmuhdur)
- GitHub Sponsors: [Sponsor MuhDur](https://github.com/sponsors/MuhDur)