Multimodal Foundation Model Scientist

Advance vision-language models at Mecka AI through multimodal research, post-training, and reinforcement learning for embodied AI systems.

This experienced Research Scientist will develop large-scale multimodal foundation models across their development lifecycle. You will research architectures, training strategies, evaluation methods, and optimization techniques. The work connects data quality, model reasoning, reliability, and real-world capabilities.

• Research and improve vision-language model architectures
• Implement supervised fine-tuning, preference optimization, and reinforcement learning
• Investigate reasoning, generalization, reliability, and training efficiency
• Design evaluations and benchmarks to identify model failure modes
• Run experiments with distributed infrastructure and large datasets

• Hands-on experience training large-scale multimodal foundation models
• Experience with Qwen, Qwen-VL, or comparable models
• Strong understanding of deep learning and optimization
• Proficiency in Python and PyTorch, JAX, or comparable frameworks
• Experience analyzing large-scale machine learning experiments

Bring multimodal model research, reinforcement learning, and evaluation expertise to Mecka AI’s embodied AI systems.