TrendMicro aimed to enhance its AI chatbot service to deliver personalized, context-aware support for enterprise customers. The chatbot needed to retain conversation history for continuity, reference company-specific knowledge at scale, and ensure that memory remained accurate, secure, and up to date. The challenge is in integrating long-term memory for organizational knowledge with short-term memory for ongoing conversations, while supporting company-wide knowledge sharing. In collaboration with the AWS team, including AWS’s Generative AI Innovation Center, TrendMicro addressed this challenge using Amazon Neptune, Amazon OpenSearch, and Amazon Bedrock, as we elaborate in this blog.

TrendMicro implemented company-wise memory in Amazon Bedrock by combining multiple AWS services. Amazon Neptune stores a company-specific knowledge graph, representing organizational relationships, processes, and data to enable precise and structured retrieval. Mem0 manages short-term conversational memory for immediate context and long-term memory for persistent knowledge across sessions. Amazon Bedrock orchestrates the AI agent workflows, integrating with both Neptune and Mem0 to retrieve and apply contextual knowledge during inference. This architecture allows the chatbot to recall relevant history, retrieve structured company knowledge, and respond with tailored, context-rich answers—helping significantly improve user experience.

The architecture begins with capturing user messages and extracting entities, relationships, and potential memories through the Claude model on Amazon Bedrock. These are then embedded with Amazon Bedrock Titan Text Embed and searched against both Amazon OpenSearch Service and Amazon Neptune. Relevant entities and memories are retrieved, and updated through the model before being re-embedded and indexed back into OpenSearch and Neptune. This closed-loop process makes sure that entity-related memories can be continuously refreshed and the knowledge graph in Neptune remains consistent with conversational insights.

When handling user queries, the system applies a similar embedding pipeline with Bedrock Titan to search across both vector embeddings in OpenSearch Service and entity triples in Neptune. The relevant memories are then reranked using Amazon Bedrock Rerank or Cohere Rerank models to make sure that the most contextually accurate information is delivered. This dual retrieval strategy provides both semantic flexibility from OpenSearch and structured precision from Neptune, enabling the chatbot to deliver highly relevant, context-aware answers.

For each AI response, the system maps sentences to the specific memories referenced, generating a memory assessment report. Users are then presented with the opportunity to approve or reject these mappings. Approved memories remain part of the knowledge base, while rejected ones are removed from both OpenSearch Service and Neptune. This makes sure that only validated and trusted knowledge persists. This human-in-the-loop mechanism strengthens trust and helps continuously improve memory accuracy and gives enterprise customers direct influence over the refinement of their AI’s knowledge.