Search-Engine-Integrated Multi-Expert Inference (SEIMEI)
SEIMEI is an open source library which allows us to create complex AI Agent system by search-engine-optimization among the agents.
The biggest feature of SEIMEI is that we can make highly reasoning AI Agent system to our own tasks through Reinforcement-Learning(RL) of search engine with super low GPU resources compared to RL of LLM.
We are still developing SEIMEI library and looking for engineers to work on this library with. Contact us!!
Fundamental Research on Search Engine
In this research, we train a proprietary search model with a proprietary training method. With the model and method, we realized more stable and solid training on search models, and achieved 16% improvement on coding task (bigcodebench/deepseek-r1) with SEIMEI.
To learn more, you can see the research result slide (canvas) from the button below.
