JapanLogistics and transport2024 – ongoing

Konoike Transport

Generative AI chat over the internal knowledge database on Amazon Bedrock

Konoike Transport's Bedrock knowledge chat passed 95 of 100 answer checks and went live in four months

WorkingVendor case studyVendor ↗

Problem

Konoike Transport's internal knowledge was scattered, which made it hard for staff to find what they needed.

Approach

A RAG chat application was built on AWS. Amazon Bedrock corrects search queries, Amazon Kendra searches the documents, and Amazon Aurora returns full records. Bedrock then writes the answer with a link to the source. The team went from start to production in four months.

Results

95件answers that passed out of 100 checked, against an acceptance target of 90% or more, per AWS Japan's blogVendor case study[1]

8倍近くdatabase search access in July 2024 compared with the previous level (nearly eight times), per AWS Japan's blogVendor case study[1]

4 か月time from start to production, per AWS Japan's blogVendor case study[1]

Takeaway

The team set a 90% pass bar for answers before launch and reported the result against it.

Sources

鴻池運輸様におけるAWS生成AI事例：Amazon Bedrockによる社内ナレッジの共通知化
AWS Japan Blog30 Sep 2024Vendor publication

Source summaries

[1]AWS Japan Blog30 Sep 2024

AWS Japan described a RAG chat application for Konoike Transport built on Bedrock, Kendra, Aurora, and S3. It reported production launch four months after start, 95 of 100 test answers passing against a 90% target, and July search access nearly eight times the earlier level.

Full details are in the original sources.

Added 2 Nov 2026 · Updated 2 Nov 2026