Home > AI Solutions > Artificial Intelligence > White Papers > Rethinking Hierarchical Text Classification: Insights from Multi-Agent Experiments with Small Language Models > Experiment 1
Experiment 1 underscored the importance of robust error-detection mechanisms to address the cascading effects of errors. Incorrect responses early in the workflow significantly impacted later outputs. The team encountered challenges with function call emulation in SLMs, but the LangGraph node system provided a viable workaround, enabling function emulation and expanding the model integration options. LangGraph also proved essential for creating customized workflows, offering flexibility to use different models at various workflow stages. Output consistency was a challenge, with SLMs often producing variable JSON structures. Validation strategies and node re-run addressed these challenges, improving accuracy but adding latency. Future efforts should focus on optimizing these mechanisms to enhance both consistency and efficiency.