AI solutions & automation
LLMs, retrieval and agents built into existing systems: assistants that qualify leads, agents that answer questions over company records, automation with tools and a human in the loop.
AI architecture & integration
I help companies take AI from pilots to production.
I design how LLMs, retrieval and agents plug into the systems a business already runs, and the security, governance and cost model that keeps them running afterwards.
- Siemens Energy · AI software architect
- Own multi-tenant AI assistant platform · running live
- 17 years in engineering
What I work on
Software & data architecture
Solution and enterprise architecture, data platforms and knowledge graphs, cloud-native infrastructure as code. The foundation AI needs before it can be useful.
Evaluation & quality
AI tested like software: simulated users, hard checks, an LLM judge and red teaming, before and after every change. Built on eleven years of test engineering.
Selected work
All projects →A multi-tenant AI assistant platform, built solo
An AI assistant that qualifies site visitors and hands usable leads to sales, for many businesses at once, without a development project per customer.
A hybrid search agent for half-structured questions
Business questions that are half structured data, half free text: an agent that answers both at once, without copying sensitive data out of its platform.
From MVP to an enterprise knowledge-graph platform
Data from disparate legacy systems in one knowledge graph, so search, analytics and new applications plug into one platform instead of each source.
Two years of an AI assistant in a beauty studio
From November 2024 an assistant on the website and WhatsApp handled about 70% of client enquiries for a small beauty studio.
A performance-critical message aggregator, delivered as a fast PoC
An extra service for a client, delivered as a proof of concept in a very short time and faster than required by an order of magnitude.
1,500 test cases behind an accelerated network stack
A vendor's network acceleration that bypasses the kernel could only ship once it behaved like the real stack. About 1,500 test cases made that provable.
About
Evaluation, not vibes
Before architecture I spent eleven years in networking and test engineering: the Linux network stack, 5G and SIP, and test frameworks covering thousands of cases. That discipline is what LLM systems need now.