PhD Position: World Models for Software Architecture Synthesis (Constructor Fabric) Constructor Knowledge Labs gGmbH in Germany

PhD opportunity in machine learning for software engineering and formal methods at Constructor Knowledge Labs gGmbH, in collaboration with Constructor Technology, on the Constructor Fabric project.

The project asks how a company’s informal knowledge can be turned into production software through a pipeline of composable capability units (“gears”), and how to build a world model over that pipeline. Rather than simply generating code, the research aims to predict the consequences of architectural decisions: total cost of ownership, unintended side effects, latency, failure behaviour, and whether a proposed composition is even admissible.

The technical core combines formal architecture representations, software graphs, DSLs, and learning-based search. The project works with gear contracts, typed attributed hypergraphs, a composition algebra for legal assemblies, and semantic projections over types, protocols, resources, and security. It also involves architecture synthesis as an inverse problem, where a specification and constraints must be mapped to a valid composition of gears, using constraint solving, Monte Carlo Tree Search, learned transition/value models, and rigorous checks that treat LLM outputs only as hypotheses.

Depending on background, the PhD may focus on formal skeletons and composition, DSL and language engineering, local transition models and pipeline surrogates, or data generation and mining of specification-to-architecture pairs from open repositories. The work is engineering-heavy and research-driven, with a strong emphasis on reproducibility, formal reasoning, and calibrated prediction.

Funding: full financial coverage through a dedicated fellowship, including a monthly stipend of €1,650, a monthly research-cost allowance of €100, and a health-insurance subsidy of €100 per month.

Eligibility highlights: MSc or equivalent in computer science, mathematics, or a mathematical natural science; strong background in probability, linear algebra, discrete mathematics, logic and formal methods, sequential decision making, and ideally language engineering/model-driven engineering. Technical English fluency is required, along with evidence of finishing work such as open-source code, publications, or competition results.

Location and timing: Bremen, Germany; full-time, 3-year temporary PhD position; start date November 2026. Applications are reviewed on a rolling basis.

How to apply: submit a CV, cover letter, code link, one-page research statement, degree certificates and transcripts, and contact information for 2–3 academic referees via the application link.