Founding engineer, AI on CAD geometry (co-founder equity)

LoopforgeLab enables machine and hardware makers make better design decisions by better we mean shorter design iterations, more sustainable products with less cost to manufacture and own.

We are building the Forge Engine is an agentic tool for manufacturing. Give it a BOM and a STEP file and it returns what the part costs to make and what the machine will cost to own over fifteen years, while there is still time to change the design. The lifecycle engine is built and runs on declared drivers. The agent already modifies geometry from natural-language instructions on our platform.

The next piece is the reason we are hiring. We need geometry reasoning that holds up on real customer CAD: feature recognition from B-rep, tolerances read per feature rather than per part, similarity against parts we have already costed, and an evaluation setup that shows whether the agent is improving or only sounding more certain. Also we need to build integrations with different CAD/ERP and PLM systems.

01What you would own

• The geometry reasoning module. Feature extraction from STEP, manufacturability signals, tolerance and surface finish handling.

• The design agent’s optimisation loop. Iterating a design against a cost or lifecycle target rather than following instructions.

• Evaluation. Benchmarks, regression suites, and the discipline to drop approaches that do not beat the baseline.

• Architecture decisions with the founder, including when LLM reasoning beats rules and when it does not.

• Reasoning Engine and live context retriever

02Who you probably are

• Strong experience with one of those: Python/Rust and C++ and real ML depth. You have built at least once in your life a geometric deep learning, GNNs, 3D representation learning modules in real production data settings and they're used by real users right now.

• You have touched mechanical engineering for real. A degree, CAE/CAD/CAM work, manufacturing process knowledge, or a PhD near feature recognition, DFM or process simulation. You know why a tight tolerance on one face costs more than a loose one on five.

• Comfortable with messy industrial data. Re-exported CAD that parses differently, BOMs with missing fields, 40 parts rather than 40,000.

• You ship. Two-week demoable checkpoints rather than six-month research tracks.

• Familiar with sustainability concepts and how do they be applied in the context of design and manufacturing

03Nice to have

• OpenCascade, PythonOCC, or commercial CAD kernel experience.

• MCP servers, agent tooling, or LLM-in-the-loop CAD work.

Co-founder-level equity with standard vesting. You would be the third engineer in the room, which means you own things outright, including the parts nobody enjoys. We plan to open source what the community builds and put a private product layer on top.

Find Jobs in Germany on Arbeitnow