Postdoc in AI-Powered Testing for Dynamic Models of Social Systems Full-time NEW

The MIT Sloan School of Management seeks a full-time Postdoctoral Associate to lead research software development and evaluation for a MGAIC-funded project led by Prof. Hazhir Rahmandad on AI-powered testing for dynamic models of social systems. The project will build an open-source pipeline using large language models, retrieval-augmented generation, and software-testing methods to generate, prioritize, implement, and continuously run quality-assurance tests for system dynamics and related simulation models.

The position is intended for a high-agency researcher-builder who can work independently to move from research design to working prototypes, evaluate the pipeline on benchmark models, and disseminate outputs through open-source software, publications, presentations, and SERC community activities.

Principal Duties and Responsibilities 

– 20% – Research design and test taxonomy. Codify model confidence-building tests into a machine-readable taxonomy; define benchmarks and test specifications for approximately a dozen smaller literature models and two complex test beds, including urban dynamics and climate/sustainability models; exercise independent judgment in scoping, prioritizing, and translating qualitative model-quality concepts into measurable checks.

– 25% – LLM/RAG test-generation pipeline. Design, implement, and iterate prompts, retrieval workflows, model parsers, and orchestration code that generate prioritized tests from natural-language problem statements and from working models and datasets; make independent technical decisions about architecture, evaluation metrics, error handling, and reproducibility.

– 25% – Executable testing and continuous integration. Generate and maintain code to execute structural, behavioral, dimensional, extreme-condition, sensitivity, and data-fit tests in model workflows; develop interfaces for Vensim, XMILE, SDEverywhere, and extensible Python/JavaScript tools; create repeatable continuous-monitoring workflows that flag regressions after model changes.

– 15% – Evaluation and validation. Compare AI-generated tests against expert-designed benchmark tests; analyze false positives, false negatives, novelty, and usefulness; conduct and document structured feedback sessions with lead modelers and domain experts; maintain reproducible datasets, analysis scripts, and results.

– 10% – Dissemination and project leadership. Prepare an open-source release, technical documentation, manuscripts, conference submissions, presentations, and final reports; participate in SERC programming and lead or organize a SERC Scholar Group during the academic year; coordinate with MIT/SERC collaborators and external modeling communities.

– 5% – General collaboration and administrative duties. Contribute to project planning, version control, issue tracking, code review, meetings, and informal mentoring of students or assistants on project-specific tasks as needed.

Other duties as needed or required.

Tagged as: Data Science