Teams at Rhoda AI

Design, build, and scale high-throughput data infrastructure for multimodal training data. Optimize storage, indexing, retrieval, throughput, and workload balancing. Implement observability, artifact versioning, and tooling to enable research and production use. Integrate VLMs and support reproducible, cost-efficient pipelines for large-scale model training.

Develop and implement post-training pipelines that adapt web-pretrained causal video models to real robot tasks. Build RL training pipelines, reward models, inverse dynamics for action execution, and evaluation frameworks. Iterate between simulation and hardware, diagnose failures, and improve generalization with few-shot adaptation and preference alignment on robot-collected demonstration data.

Optimize inference performance of large multimodal foundation models across cloud and on-robot targets. Diagnose bottlenecks, apply quantization/pruning/distillation, tune kernels (CUDA/Triton), build benchmarking and regression detection, and translate research models into deployment-ready implementations.