QUICK FACTS:

<aside> 🌎

LOCATION: Munich, Germany - hybrid

</aside>

<aside> 🛠

OUR TECH STACK: ”Classic” Software Stack: TypeScript, Next.js, Python, PostgreSQL, Terraform ”Modern” AI Stack: Leading LLM models, observability/evaluation tooling (Langsmith), graph RAG & vector databases

</aside>

<aside> 💼

EMPLOYMENT TYPE: Full-Time

</aside>

<aside> 📅

START DATE: asap

</aside>

🚀 What WE do at Tendos

We are Tendos, a Munich-based startup, offering agentic workforce for white-collar jobs in construction. Our AI-first solution streamlines and automates tendering and quoting related workflows in the construction industry. Tendos is freeing inside sales teams from tedious manual tasks while unlocking new revenue potential. Take the unique opportunity and join one of the hottest early-stage start-ups to redefine workflows with AI in the biggest industry in the world!

🚀 The role

Make the agents we already have better, and build the ones we don't. There is a working system in production: retrieval over a large catalogue corpus, extraction, classification, matching, and a rule graph that learns from corrections made by the people who own the commercial consequence of a mistake. It works, and it is nowhere near what it could be. Every increment in what these agents can do reliably is another piece of somebody's week that stops being manual.

That is the whole job, and it is more specific than it sounds. Most of the distance between what our agents do today and what they could do is not a model problem. It is knowing which failure is worth chasing, being able to tell a retrieval failure from a selection failure before you start fixing either, and making a change you can prove was an improvement rather than a different set of mistakes.

🥇 What YOU will do at Tendos

• Move production quality, measurably. Error analysis, a hypothesis, a change, and evidence it was an improvement. The skill is in the loop, not in the build.
• Own the evaluation harness. Datasets pinned to an event store so a dataset is a reproducible pointer rather than a stale export, and a failing trace that narrows to a single line item.
• Decide which agent capability is worth having next - and when the honest answer is not yet, or not at this price. Almost anything here can be automated eventually; the judgement is what it costs to build and to run.
• Ship your own work to production. Containers, serving, an environment you do not control.
• Keep us honest on cost and latency. We have no cluster and no GPU fleet, and we are not going to pretend otherwise.

💪🏼 What we are looking for

We are looking for a Senior AI Engineer to make our agents better and to own the harness they are measured on. You will actively shape the growth path and tech setup of our organization to become the market leader for AI Agents in the construction space.

• Five or more years building software, at least two of them on LLM or machine-learning systems that ran in production and had real users.
• You have made a production LLM system measurably better, and can say how you knew. The metric that moved, the harness it moved on, and why you trusted it. Whether you got there through retrieval, prompting, decomposition, fine-tuning or deleting a step matters far less than the evidence.