Intern, Research Foundational Models

Job Requisition ID #

25WD92303

Position Overview
We are seeking a research intern to explore fundamental challenges in geometry, design understanding, and relative spatial reasoning for vision-language models (VLMs). While modern VLMs have shown strong performance on captioning, semantic understanding, and segmentation, they continue to struggle with geometric reasoning, layout understanding, and precise relative positioning-capabilities that are critical for design, engineering, and creation workflows.
During this internship, you will work closely with research mentors to investigate new modeling and training paradigms that move beyond one-shot visual reasoning. The project will focus on approaches such as reinforcement learning, test-time computation, and thinking with images, where models iteratively attend to visual evidence, reason over intermediate representations, and verify hypotheses through visual feedback. The goal is to advance state-of-the-art methods for spatially grounded reasoning and contribute insights relevant to both the research community and Autodesk's long-term vision for intelligent design tools.

• Define and execute a research project focused on geometric reasoning, spatial understanding, and layout awareness in vision-language models
• Conduct literature reviews to identify limitations of existing VLMs and relevant prior work in multimodal reasoning and reinforcement learning
• Design and implement novel training or inference strategies using reinforcement learning, test-time computation, or iterative visual reasoning
• Develop model architectures, training pipelines, and evaluation benchmarks for spatial and geometric tasks
• Run large-scale experiments, analyze results, and iterate on model designs based on empirical findings
• Compare proposed approaches against strong baselines and state-of-the-art methods
• Collaborate closely with research mentors and other researchers, sharing progress and incorporating feedback
• Author a research paper suitable for submission to a top-tier machine learning or computer vision conference
• Present research results internally at Autodesk and externally at academic venues

• Currently enrolled in a PhD program in Computer Science, Machine Learning, Computer Vision, or a closely related field
• Must have at least one academic remaining semester post internship
• Strong publication record in top-tier ML or vision conferences (e.g., ICML, NeurIPS, ICLR, CVPR, ICCV, ECCV)
• Hands-on experience training vision-language models and reinforcement learning algorithms
• Strong implementation skills using modern deep learning frameworks (e.g., PyTorch, TRL, Ray)
• Solid background in machine learning fundamentals and experimental research methodology
• Ability to work independently on open-ended research problems and communicate results clearly

• Experience with multimodal or embodied reasoning, test-time optimization, or iterative inference methods
• Familiarity with geometric vision, spatial reasoning benchmarks, or synthetic visual datasets
• Experience scaling experiments on distributed systems or large compute clusters
• Strong written and verbal communication skills