What challenges did Trend Micro face before AI implementation?

Improve an AI chatbot so it can deliver personalized, context-aware support for enterprise customers. Retain long-term, company-specific organizational knowledge while keeping memory secure, accurate, and up to date.

What was the impact of the AI solution?

Expected improved answer quality and more accurate, verifiable, organization-specific responses at scale. Foundation for continuously adapting to evolving organizational knowledge; work is under evaluation and tuning.

Source

Scaled productionEvidence: Low40/100

Trend Micro builds company-wise persistent memory for its Bedrock chatbot using Amazon Neptune

Use case typeCustomer support automationUpdated Apr 22, 2026

Trend Micro built the Trend’s Companion chatbot to provide enterprise customers with natural, conversational interactions that remain personalized and context-aware over multiple sessions. The solution combines Amazon Bedrock, Amazon Neptune, Amazon OpenSearch Service, Titan Text Embed, reranking, and Mem0 to create persistent company-specific memory and a human-in-the-loop approve/reject workflow.

Organization: Trend Micro
Industry: Tech & Comms
Location: Japan
Published: April 2026

Reported outcomes

Strategic outcomes

Customer experience & trustPersonalized, context-aware enterprise supportSpeed & agilityPersistent company-specific memoryRisk & complianceValidated memory retention with human oversight

Why do we believe this deployment?Customer identity, provider attribution, maturity, and source checks

Customer: Trend Micro
Provider: AWS
Maturity: Scaled Production
Linked source: AWS Machine Learning Blog

Expected improved answer quality and more accurate, verifiable, organization-specific responses at scale

Customer identity supportedSource describes one deploymentMaturity supported

Primary read

Use case focus

Showing 2 of 2

1Customer support automation
2Knowledge management

Improve an AI chatbot so it can deliver personalized, context-aware support for enterprise customers.
Retain long-term, company-specific organizational knowledge while keeping memory secure, accurate, and up to date.

Extract entities, relationships, and candidate memories from user messages.
Embed and index them into OpenSearch and Neptune; retrieve from both sources, rerank results, and use Bedrock to orchestrate grounded responses.
Add a human-in-the-loop approve/reject loop that removes rejected memories from OpenSearch and Neptune.

Technologies

Amazon Bedrock Amazon Neptune Amazon OpenSearch Service Amazon Bedrock Titan Text Embed Bedrock Rerank Cohere Rerank Mem0

Expected improved answer quality and more accurate, verifiable, organization-specific responses at scale.
Foundation for continuously adapting to evolving organizational knowledge; work is under evaluation and tuning.

Architecture

Trend Micro’s chatbot architecture uses Amazon Bedrock to orchestrate agent workflows, Amazon Neptune to store company-specific knowledge graphs, Amazon OpenSearch Service for vector search, Amazon Bedrock Titan Text Embed for embeddings, Bedrock Rerank/Cohere Rerank for relevance ordering, and Mem0 to manage short- and long-term memory. The system extracts entities and candidate memories from user messages, indexes them, retrieves from both graph and vector stores at query time, reranks the results, and applies a human-in-the-loop approve/reject loop to keep persisted memory validated.

Sources & evidence1

Evidence: Low40/100Evidence strength