| Task decomposition, tool use, context & memory, workflow orchestration and result validation. | Data construction, training strategies and task-specific adaptation for practical applications. |
| Model structure, inference efficiency and deployment compatibility for stable, efficient systems. | Compression, quantization, ONNX / TensorRT acceleration, serving and edge devices. |
vLLM |
SGLang |
Whisper |
pytest |
Agent Development Kit Google harness2 |
DeepSeek Harness |
Reproduction notes, fixes and verification logs are archived in open-source-contributions.