> Markdown version of [/jobs/ext/4545799-machine-learning-engineer](https://www.wearedevelopers.com/jobs/ext/4545799-machine-learning-engineer). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md).

---

# Machine Learning Engineer

- **Company:** Deft AI, Inc.
- **Location:** San Francisco, CA, United States
- **Experience:** Experienced
- **Salary:** $150,000 - $200,000
- **Contract:** Permanent contract
- **Skills:** Machine Learning, Tensorflow, Reinforcement Learning, Pytorch, Information Technology, Machine Learning Operations
- **Published:** November 3, 2026
- **Apply:** https://www.careerbuilder.com/job-details/machine-learning-engineer-san-francisco-ca--d479709e-3b79-4276-b464-0a365a30a54f

## About the Role

Machine Learning, Constraint Satisfaction, Benchmarking, Data Collection, Vehicle Fleets, Use Cases, Computer Science, Machine Learning, Robotics, Reinforcement Learning, PhD or MS degree in Computer Science, Machine Learning, Robotics, or equivalent technical discipline
Deep expertise in machine learning fundamentals, reinforcement learning, and associated
frameworks (PyTorch, TensorFlow, Ray, etc.)
3+ years of proven track record developing and deploying ML systems from research through production implementation
Hands-on experience with model lifecycle management including training, deployment, and maintenance in production settings

Preferred Qualifications
Authored or co-authored peer-reviewed publications in robotics or related fields
Hands-on experience designing and implementing bimanual manipulation tech stacks
with imitation learning or RL-based methods
Background in real-time ML inference systems, simulation-to-reality transfer, or advanced
reinforcement learning implementations

## Description

Design and develop a modular robot autonomy stack that composes Vision-Language-Action (VLA) models with purpose-built modules to enable grasping and dexterous behaviors in unstructured environments
Develop action refinement and safety layers that post-process VLA outputs - constraint satisfaction, collision and force guards, smoothing, and runtime monitors for safety-critical deployment
Architect clean interfaces and abstractions around base VLA models so they can be swapped, benchmarked, and upgraded as the SOTA evolves - keeping the stack model-agnostic
Design and maintain robust data collection and curation pipelines for production robot fleets
Build reliable, high-speed robot autonomy software stack optimized for inference
performance
Advance SOTA dexterous manipulation architecture through novel methodologies while bridging theory & practice-real customer use-cases with clear success criteria.