AI BriefingImproving Memory Retrieval Performance: New Computer Achieves 50% Recall Boost with LangSmith
Key point
New Computer leveraged LangSmith to improve its agent memory system's recall by 50% and precision by 40%.
Details
New Computer, the developer of the personal AI 'Dot', is building a long-term memory system that learns from users' linguistic and behavioral cues. Moving away from conventional static RAG approaches, they introduced an Agentic Memory system that structures information from the memory creation stage to improve retrieval efficiency.
New Computer used LangSmith to test and optimize various retrieval methodologies. The key results are as follows.
- 50% improvement in Recall
- 40% improvement in Precision
To protect privacy, they used LLM-generated synthetic data, and ran multiple techniques—including Semantic Search, BM25, and Meta-field Filter—in parallel to find the optimal combination. LangSmith's SDK and Experiments UI played a key role in efficiently managing this complex experimentation process and validating the impact of prompt changes on the overall system.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.