Research Scientist, Computer Vision
Attentive.ai
·Aug 31, 2026
- Salary
- $170k – $275k
- Location
- San Jose, United States
- Workplace
- Remote
- Department
- Education
- Education
- PhD
- Visa
- Sponsored
- Source
- Keka
Description
- Own research problems end to end: framing, data, experiments, ablations, and the model that ships
- Advance deep learning models for object detection, semantic segmentation, and structured extraction on vector and raster construction drawings
- Build multimodal systems combining visual layout, geometry, and text
- Design evaluation frameworks to continuously improve the model
- Stay current with computer vision and ML research and evaluate what's relevant to construction-industry problems
- 2–8 years of applied machine learning or AI research experience with a strong focus on computer vision and deep learning. We'll calibrate level and compensation to what you've built.
- Depth in at least one of: object detection, image segmentation, image processing, OCR and layout analysis, self-supervised or representation learning, or multimodal / vision-language models
- Experience training and evaluating neural networks at scale, and at least one model you've taken to production
- Strong Python and PyTorch, across the modeling, training loop and the data pipeline
- Practical GPU and distributed training experience
- Judgment about data: what to label, what to synthesize, what to discard
- The ability to read current research critically and say what's worth trying
- Master's or PhD in Computer Science, Machine Learning, Electrical Engineering, Mathematics, or a related field
- Familiarity with vector graphics and CAD formats (SVG, DXF, DWG, IFC)
- LLM and NLP experience like structured extraction, retrieval over long documents
- Model optimization: quantization, pruning, knowledge distillation
- Publications or open-source contributions in computer vision or deep learning
- Experience with cloud environments (GCP, AWS, or Azure)
- Empirical over theoretical — we'd rather run the experiment than argue about it
- Truth-seeking, including about our own results. A negative result reported early is worth more than a positive one defended late
- Research is judged by whether it holds up in front of 1,200 contractors, not by whether it's clever
- Honest feedback, given directly