4+ years of experience in training and deploying real-world AI solutions
Deep understanding of model architectures, training pipelines, and evaluation methodologies
Creative and technically adept in exploring various algorithmic approaches
Proven ability to convert unstructured data into usable training sets
Strong background in algorithms and data structures relevant to spatial and generative problems
Hands-on experience with procedural generation techniques related to 3D geometry
Expert proficiency in Python, with knowledge of static typing methods preferred (type hints, MyPy, Pydantic)
Bonus: Frontend development skills (React/Three.js) are a plus.
Responsibilities
Own the full lifecycle of novel spatial model development
Define inference problems and structure training data from image sources
Experiment with ML and procedural generation to achieve efficient outcomes
Design and train novel deep learning models for inferring room layouts and object placements
Develop and optimize procedural generation algorithms for diverse room layouts
Benchmark model performance across accuracy, speed, and generalization
Integrate AI models into the product stack with full-stack engineers to enhance 3D design interactions.
Benefits
Opportunity to lead and shape foundational AI technology efforts
Work in a dynamic environment adapting various algorithmic approaches
Usable outputs integrated into interactive 3D environments
Potential to explore cutting-edge ML and spatial modeling techniques
Flexible problem-solving challenges that encourage innovative thinking.
Full Job Description
We are looking for a Lead AI & Algorithms Engineer to own the full lifecycle of novel spatial model development; defining the right framing of complex inference problems, structuring image, video, and vectorized floorplan data into usable training formats, training and evaluating models, and deploying performant systems into our production stack. You'll be responsible for developing and advancing our diffusion models and 3D generative models that power Dream's core spatial intelligence.
Upcoming Challenges
Create training datasets from imperfect sources (LiDAR Scans, scraped construction docs)
Combining heuristics, labeling tools, and clever preprocessing (e.g. object detection models) to generate high-quality data at scale.
Design and train novel spatial deep learning models that intelligently infer room layouts, surface boundaries, and object placement from partial inputs - balancing architectural plausibility with flexibility and precision.
Explore hybrid approaches that combine classical geometry algorithms with learned models.
Benchmark and iterate on model performance across multiple axes - accuracy, speed, generalization
Integrate AI models directly into the product stack, working closely with full-stack engineers to ensure outputs are usable in the interactive 3D design environment.
What we Require
4+ Years of experience training and deploying real-world AI . Deep understanding of model architectures, training pipelines, and evaluation methodologies.
Creativity and technical depth to explore, evaluate, and iterate on different algorithmic approaches.
Scrappiness and speed in turning unstructured or incomplete data into usable training sets, and in getting models built and tested under real constraints.
Strong background in algorithms and data structures, with the ability to design performant solutions to spatial and generative problems.
Hands-on experience with procedural generation techniques, especially in the context of 3D geometry, modeling, or simulation.
Expert proficiency in Python, with a strong preference for experience using static typing (e.g., with type hints, MyPy, or Pydantic).
