Build reliable multi-agent AI systems for real-estate workflows. Responsibilities include developing Python delivery pipelines, context assembly, tool orchestration, verification, report generation, data extraction, evaluation harnesses, observability, regression datasets, and production-quality reliability. The intern will collaborate with architects, clients, and domain experts while working with regulatory, financial, market, and geospatial data.

Conduct research on discrete diffusion models for compliant architectural floorplan generation. Design and evaluate falsifiable ablation studies, improve model architecture, loss functions, conditioning, sampling, and distributed multi-GPU training. Build computer vision pipelines to transform plans, PDFs, and scans into structured training data. Collaborate with AI engineers, researchers, and architects to turn foundational research into deployable design tools.

Own the end-to-end data strategy for a foundation model generating building geometric graphs. Build pipelines converting floorplans into canonical graphs, generate synthetic data, curate and validate datasets, design pretraining mixtures, run large-scale training and ablations, and develop evaluation infrastructure. The role requires hands-on Python engineering, distributed data processing, data quality systems, and experience personally building datasets for pretraining.