Founding AI Engineer | High Growth Start-Up | Lisbon or Remote

Planeco Building; Deutschland; À distance; EUR 120 000 - 180 000; Plein temps

Résumé du poste

Planeco Building seeks a Founding AI Engineer to build an AI product that automates planning for construction projects. You will own the agent runtime, shape architecture, and work with domain experts to scale a fast-growing startup in Europe, with options for Lisbon office or remote work.

You'll collaborate directly with the founders and Product Lead, shipping features weekly while establishing guardrails for reliability, security, and performance as the product matures.

Qualifications

• Built the whole body of an agent: retrieval and RAG, MCP, tool calling, evals.
• House a portfolio of shipped projects in a product setting.

Responsabilités

• Build and scale the agent runtime: core loops, state machinery, deterministic fallback paths.
• Manage context and token mechanics: dynamic context assembly, memory pruning, token budgeting, compaction.
• Ship tool executions and integrations: performant tool-calling handlers and MCP clients/servers.
• Enforce schema and output integrity: robust structured output parsers for real traffic.
• Architect state and human-in-the-loop systems: durable state persistence and checkpointing.

Description du poste

Founding AI Engineer | High Growth Start-Up | Lisbon or Remote

We’re turning around one of Germany’s largest industries. As the first in the EU, we’re building an AI product that takes over the entire planning process for construction projects and driving the transformation of a whole sector with it. More than 3,000 projects a year give us the data to build our AI on.

Who we are

Planeco Building does end-to-end real estate projects on autopilot.

We’re building the infrastructure that digitises planning and permitting for real estate projects of every size, end-to-end: faster, cheaper, scalable. We combine our own architects with an external partner network and a tech and automation layer that streamlines or fully replaces many manual processes.

We build everything: from carports and Airbnb conversions to large apartment buildings and padel halls.

• Marketleader in Germany
• 15–20% MoM as a bootstrapped venture
• Profitable with a run rate over €10M
• 40-person team in Lisbon and remote

What we offer

• Real greenfield: No legacy system, no inherited codebase. You set the standards for how this product gets built.
• A test bed with 3,000 projects a year: Real cases with unambiguous outcomes. What you ship shows up the same week in hours per project and turnaround time.
• Domain experts in-house: Architects and structural engineers sit with us. Ground truth comes from a conversation, not a research rabbit hole.
• Budget for models and tools: API costs, Cursor, Claude Code, hardware of your choice. Three-figure inference costs aren’t a discussion when there’s an experiment behind them.
• Access instead of approvals: Direct work with the founders and the Product Lead. Decisions happen in conversation, prod access from week one.
• Bootstrapped and profitable: No board theatre, no fundraising quarters. The runway comes from customers.
• Lisbon or remote: Work from our office in central Lisbon, where our chef cooks lunch and dinner three times a week, or remote from anywhere in Europe.

What you’ll do daily

• Build and scale the agent runtime: Own the core loops, state machinery and deterministic fallback paths feeding into complex backend workflows.
• Manage context and token mechanics: Dynamic context assembly, memory pruning, token budgeting, compaction.
• Ship tool executions and integrations: Performant tool-calling handlers and custom MCP clients and servers, so agents can reach internal APIs, databases and third-party systems.
• Enforce schema and output integrity: Robust structured output parsers that hold under real traffic.
• Architect state and human-in-the-loop systems: Durable state persistence and checkpointing, so agents can pause for approval, survive crashes and resume cleanly.
• Implement guardrails and security: Prompt-injection defences, PII masking, permission gates before tool execution, token budget caps.
• Trace trajectories and run evals: Telemetry across multi-step execution graphs, latency and cost tracking, continuous regression evals.

You’ll fit in if you...

• have built the whole body of an agent, not one limb: retrieval and RAG (vector and semantic search), MCP, tool calling, evals.
• are at home driving models from the CLI and treat Claude Code or Cursor as daily tools.
• have a portfolio we can look at. Side projects count as much as professional work.
• have shipped inside a product team, not only solo, we want daily standups and people who know what everyone else is working on.
• communicate clearly. Nothing disappears into a branch for three weeks.
• hold a high quality bar from the start. Our output ends up in permit applications filed with German authorities. It can’t be “mostly right.”

Years of experience and job titles aren’t criteria for us. Having managed other engineers isn’t either. What counts is what you’ve shipped and how you think about it.