Actify AI is focused on building a new generation of AI agents capable of performing real computer tasks end-to-end. They are seeking a Machine Learning Engineer to optimize and scale VLA-based computer-use models, working on reinforcement learning and model deployment for efficient real-world applications.
Responsibilities:
- Optimize VLA/computer-use models for long-horizon desktop and browser automation tasks
- Design and implement reinforcement learning pipelines for agent behavior improvement
- Improve model reliability, task completion rate, action accuracy, and recovery from errors
- Build evaluation systems for computer-use tasks, including benchmarks, task suites, and failure analysis
- Optimize inference performance, serving cost, latency, and throughput
- Deploy and maintain production ML serving systems
- Work closely with product and engineering teams to turn model improvements into user-facing features
- Analyze model failures and develop targeted training, prompting, fine-tuning, or RL strategies to improve performance
Requirements:
- 3+ years of experience in machine learning, reinforcement learning, model optimization, or ML infrastructure
- Strong experience with reinforcement learning, imitation learning, offline RL, or agent training
- Experience with inference optimization and serving large models in production
- Familiarity with transformers, multimodal models, VLMs, VLAs, or computer-use agents
- Strong Python and PyTorch experience
- Experience with model evaluation, data pipelines, and experiment tracking
- Ability to debug complex model behavior and turn ambiguous failures into concrete improvements
- Comfortable working in a fast-moving startup environment
- Experience with browser automation, desktop automation, robotics, or embodied AI
- Experience with RLHF, RLAIF, policy optimization, reward modeling, or trajectory-based training
- Experience with quantization, vLLM, TensorRT, ONNX, CUDA, Triton, or distributed inference
- Experience building agent benchmarks or task-completion evaluation systems
- Experience fine-tuning VLMs, multimodal LLMs, or action-prediction models
- Experience with data collection from real user workflows or synthetic task generation