D
Drafted Tech Stack
AI-powered home design generation with iterative user controls
Drafted is a 2–10-person AI design tool for home visualization, built on PyTorch + Triton + CUDA with React + Three.js on the frontend. The stack reveals a compute-intensive generative ML operation: they're investing heavily in inference optimization (model distillation, vector outputs, reinforcement learning) while scaling labeling and generation pipelines—classic growing pains for a model-heavy product. All six open roles are senior-level engineers and designers, indicating they're hiring for depth, not headcount.
Tech Stack 18 technologies
Core StackFigma PyTorch React Tailwind CSS Python TypeScript Rust AWS Cloudflare Triton CUDA SWR Three.js Railway SQL PostHog Pinterest Instagram
What Drafted Is Building
◆Challenges
- Reducing inference cost
- Speeding up labeling
- Managing capacity-constrained pipelines
- Preventing duplicate work
- Handling high variance complexity
- Strengthening product-led growth strategy
- Identifying growth opportunities
- Embedding growth loops
▲Active Projects
- Design, implement, & iterate on editing generated plans using our existing machine learning model capabilities.
- Design, implement, and iterate on how users guide the output home design with new model controls (i.e. multi-floor, exterior design, etc).
- Design, implement, and iterate on the existing user interface to reduce friction and increase the number of users creating a home they love.
- Teach our ml model how to generate multi-story outputs
- Lower the cost of inference by distilling our model
- Speed up our data labeling processes with computer assisted labeling
- Teach ml model to handle multiple layers of context and constraints
- Add direct vector output to the model
- Use reinforcement learning to increase model accuracy
- Iterate on our current generation pipeline to make it more scalable and efficient
Hiring Activity
6 roles · 1 in 30d
Department
Seniority
Company intelligence
