How Dosu Used LangSmith to Achieve a 30% Accuracy Improvement with No Prompt Engineering

Dosu, an engineering teammate that automates work and protects engineers from unnecessary interruptions, uses LangSmith to improve its performance without prompt engineering. Instead of relying on prompt engineering or fine-tuning models, Dosu employs continual in-context learning by collecting feedback from users and transforming it into few-shot examples. This technique allows Dosu to adapt to organizational changes over time and has led to significant improvements in its auto-labeling accuracy. The company is actively exploring other ways to integrate continual learning into its various tasks.