Research Intern, Special Projects at Higharc in Remote

Summary: Higharc is a VC-backed startup revolutionizing home design and construction. They are seeking Research Interns (PhD) to work on vision-language models and AI systems aimed at enhancing home design workflows through natural-language instructions. Responsibilities: - Build data pipelines that extract and curate structured datasets from Higharc's existing design data - Design and implement models (or adapt existing ones) for vision-language understanding, structured output generation, and agentic editing of home designs - Implement and execute training, fine-tuning, and post-training pipelines, including preference optimization and RLHF-style methods - Develop evaluation suites and error taxonomies for targeted tasks, including layout validity, instruction-following accuracy, and edit correctness - Run rigorous ablations and scaling experiments; track results and maintain reproducibility and research hygiene - Prototype product-facing integrations that demonstrate end-to-end AI-assisted design workflows - Document findings and deliver results through technical reports, demos, and a publication-track draft Required Qualifications: - Active enrollment in a PhD program in Computer Science, Machine Learning, or a related field at a U.S. institution - Strong Python skills and hands-on experience building deep learning training workflows (PyTorch preferred) - Solid understanding of at least two of: computer vision, transformers, LLMs, representation learning, and ML experimentation practices - Demonstrated research ability through publications, preprints, open-source research code, or equivalent evidence of research impact - Comfort working in research/engineering stacks such as Roboflow, Modal, and W&B Preferred Qualifications: - Experience with vision-language models, multimodal foundation models, or grounded document understanding - Experience building or fine-tuning agentic LLM systems that reason over structured state and produce structured edits - Familiarity with post-training techniques: RLHF, DPO, or other preference optimization methods - Experience with multi-GPU training and large-scale experimentation (Modal or similar) - Familiarity with AEC data and workflows, including CAD/BIM concepts, floor-plan understanding, or domain-specific labeling Required Skills: Python, Deep learning training workflows, PyTorch, Computer vision, Transformers, Large language models, Representation learning, Machine learning experimentation, Vision-language models, Multimodal foundation models, Grounded document understanding, Agentic LLM systems, Post-training techniques, Reinforcement learning from human feedback, Direct preference optimization, Multi-GPU training, Large-scale experimentation, Roboflow, Modal, Weights & Biases, AEC data, workflows, CAD/BIM concepts, Floor-plan understanding, Domain-specific labeling, Research-oriented mindset, Autonomy Benefits: Comprehensive medical, dental, and vision coverage, Unlimited PTO, Meaningful maternity/paternity leave, Short and long-term disability plans, 401K, A stipend to create the ideal home office