Research Engineer

Responsibilities

• Design and conduct original research in spatial reasoning for residential construction, developing models that understand architectural structure, spatial relationships between building components, and the geometric constraints that govern how homes can be built.
• Design end-to-end LLM training and fine-tuning pipelines tailored to construction domains, developing domain-specific pre-training and SFT datasets from architectural drawings, building codes, product specifications, and construction documentation.
• Apply computer vision techniques to extract semantic structure from architectural inputs and develop multimodal training pipelines that link visual and textual representations of building data.
• Develop RL-based approaches for generative architectural design and architect RAG systems that ground LLM outputs in authoritative construction knowledge bases.
• Define construction-domain benchmarks, build ground-truth datasets, and establish eval pipelines that measure model performance against real homebuilding tasks.
• Research emerging foundation models and parameter-efficient fine-tuning approaches (LoRA, QLoRA, adapter layers) to inform Higharc's model strategy.
• Deliver research outputs ready for integration, including documented model artifacts, serving-ready pipelines, and clear API contracts.

Requirements

• 5+ years of professional software engineering experience, with 2+ years directly on LLM development, fine-tuning, or applied ML systems in production.
• Hands-on experience with model training frameworks (HuggingFace Transformers, PyTorch) and fine-tuning workflows on domain-specific corpora.
• Practical experience with RAG architectures, vector databases (Pinecone, Weaviate, pgvector), and embedding model selection.
• Experience with 3D modeling, computational geometry, or computer graphics in a research or production context.
• Strong proficiency in Python across the ML ecosystem; React/TypeScript proficiency for integration work.
• Master's degree in Computer Science, Machine Learning, AI, or a closely related field.

Nice-to-Haves

• Ph.D., particularly for candidates focused on model research and pre-training.
• Open-source contributions or published research.
• Familiarity with BIM tools (Revit, AutoCAD) or BIM data formats (IFC, gbXML).
• Experience with Three.js or equivalent 3D frameworks.