Founding AI Engineer

The construction industry loses $1.6 trillion annually to inefficiencies — not because people are careless, but because critical knowledge is trapped in PDFs, transcripts, and scattered systems. Every new project starts from scratch.

We're building the opposite: an AI co-worker that extends the Projektsteuerer (the person who runs complex construction projects), while every project makes our system measurably better. The moat isn't the model — it's the compounding project memory no competitor can replicate.

Pre-seed led by Realyze Ventures — LPs include Zech and other major European construction groups. Co-investors: D11Z (the family office behind Aleph Alpha) and the CDTM Venture Fund (backed by 300+ CDTM alumni including founders of Personio, Alasco and the Technical Director of DeepMind).

Our software is running today on a major autobahn construction program and an S-Bahn transit program — multi-year timelines, hundreds of thousands of pages of specs, protocols, and communications. Real consequences when we get it wrong.

• **1. Agent harness engineering for construction documents
  **One summary doesn't fit all. "Structural risk" in an RFI means something different than in a cost review, which means something different in a schedule reconciliation. We build multi-agent harnesses — specialized extraction, reasoning, evaluation stages — that route 400-page tender documents and protocol archives to the right pipeline with the right context.
  If you've read Anthropic's or LangChain's writing on agent harnesses and thought "yes, that's the hard part of shipping production agents" — that's this job.

• **2. Project memory as a compounding moat
  **We started with meeting transcripts. We're building a decision graph that grows with every project — tracking not just what was decided, but why, by whom, against which alternatives, and with what outcome. That graph feeds the next project. Every closed workflow makes the next one faster.
  This is the operational-continuity layer no ConTech player is building. We want someone who gets excited that the hard part here is not retrieval — it's deciding what signal to keep.

• **3. Context compression for 5 year projects
  **What should the system remember? Forget? Surface at which decision point? There's no clean top-k answer when a project spans 5 years and touches 50 stakeholders. This is an open research problem we're solving in production — and we'd rather hire someone who reads papers than someone who installs libraries.

• Product sense: you reason from user pain → solution → measurable outcome. You can talk to non-technical customers and understand their workflows.
• Velocity + craft: prototype fast, measure everything, iterate on real feedback. But you care about reliability because the downside of wrong is real in construction.
• Comfort with the unknown: many of these problems don't have Stack Overflow answers. You read papers, prototype, and compare approaches.
• Bonus: strong open-source work, previous early-employee startup experience, domain depth in document understanding or agent systems, or shipped systems that replaced hours of human labor.
• Level is opportunistic: We've seen brilliant new grads outperform staff engineers and vice versa. If you're exceptional, we'll find the right scope.
• Language: We work in English. German is nice-to-have for customer conversations but not required — we have native speakers for that.