Lukas Hartrumpf · Applied AI Architect · Nürnberg, Germany
Four things
at once.
That is what agentic engineering actually buys you. Below is the career most CVs flatten into a list, drawn the way it happened.
Concurrency
A timeline hides the point. These bars overlap on purpose.
Hover or tap a bar
Employment
Where the salary came from
Client engagement
What I was actually delivering, client names withheld
Independent
My own projects, outside client work
May 2025 to now
nerdware GmbH
Business Unit Lead, Engineering Intelligence
Leading people was already the job. What changed with Engineering Intelligence, the unit I run, is that I now carry several client projects at once and everything commercial around them: proposals, pricing, contract negotiation, follow-on business and the financial side of the portfolio, planned directly with the managing directors. Several of the unit's engagements are ones I won myself. Five engineers report to me today.
- Pre-sales
- Proposals and pricing
- Contract negotiation
- Portfolio and P&L
- Architecture sign-off
- Line management
Case in focus
A production system that largely wrote itself.
Since November 2025 I have been building a seminar management portal for a healthcare training institute. A solo build. It invoices real customers.
The volume is not the interesting part. The intake is. Requirements arrive the way they always do, as email from someone who is not thinking in tickets. A command reads the mailbox, turns each message into a backlog item, and an agent loop works the backlog off and writes up what it did. My job shrank to reading those reports and deciding what happens next.
The domain has real edges too: invoicing, dunning, statutory withdrawal periods, cancellation logic. Exactly the rules where a confident wrong answer arrives with an invoice attached.
- 1,564
- Commits on one agent-built production system in nine months
- 8
- Engineers led at peak, five reporting to me today
- 4
- Client engagements running in parallel
- 300,000
- Daily users on a platform I have been building since 2023
What I bring
Agent work is easy to demo and hard to keep running. The second half is the job.
Agent harness engineering
Making a coding agent reliable inside one specific codebase, which is a different job from using one.
- Claude Code
- Claude API
- Agent skills and commands
- MCP servers
- Multi-agent orchestration
- Evaluation frameworks
- Prompt engineering
- LiteLLM
- AWS Bedrock
- RAG and vector stores
Systems that have to stay up
Fifteen years of the unglamorous half: integration, compliance, migrations, and the on-call pager.
- Event-driven architecture
- Serverless
- Microservices
- Enterprise integration
- AWS
- Terraform
- Docker
- Kubernetes
- CI/CD
Languages I write daily
Not a list of everything I have ever touched. A list of what I would start a project in tomorrow.
- Python
- TypeScript
- Node.js
- Java
- SQL
Available for
Small, remote, and clearly scoped.
I have a full-time job I like, so the shape that works for me is narrow: a defined outcome, delivered asynchronously, reviewed on a call every few weeks. That constraint suits agent-heavy work better than it sounds, because the bottleneck is rarely hours at a keyboard.
Good fit
- An agent harness built for your codebase, not in general
- An AI feature you prototyped and cannot get to production
- An architecture review that ends in a written verdict
- A team that wants to work this way and needs someone who already does
Not a fit
- Forty hours a week, six months, on site
- Anything billed by the hour, where working faster costs me money
- Staffing a seat rather than delivering a result
lukas.faatz@gmail.com · Nürnberg, Germany · Native German, fluent English
On paper
- AWS Certified Solutions ArchitectCertification
- iSAQB Certified Professional for Software ArchitectureCertification
- B.Sc. Business InformaticsGeorg Simon Ohm Hochschule, Nürnberg, 2019
- Fachinformatiker AnwendungsentwicklungVocational degree in software development, 2009 to 2012