Case study · Manufacturing · Document intelligence
Intelligent document search streamlines machinery maintenance for a consumer electronics giant.
HVACR manufacturing · Global
Technicians lost one to two hours a day searching disorganised manuals. We put a Kendra-backed assistant in front of all of it.
70%
Less time spent on manual searches
50%
Faster information retrieval
1-2 hrs
Per technician, per day, recovered
The opportunity
The answer existed. Finding it took an hour of the working day.
This world leader in heating, air-conditioning and refrigeration maintains technical documentation across user manuals, maintenance guides, schematics and troubleshooting procedures. For a technician working on complex machinery, that documentation is the roadmap.
But it was disorganised and spread across systems, so technicians averaged one to two hours a day simply searching for information. Issue resolution took longer as a direct result, and access to critical troubleshooting material was inefficient exactly when it mattered most.
Documentation was disorganised
Manuals, guides, schematics and procedures had no unified index.
1-2 hours a day searching
Technicians spent that long on average locating information.
Content spread across systems
Material sat in S3, SharePoint, Confluence and websites separately.
Longer issue resolution
Slow access to troubleshooting material extended every fix.
The impact
Ask a question, get the passage-not a folder to search.
One index across four sources
S3, SharePoint, Confluence and website connectors ingest into a single Amazon Kendra index.
Answers, not results
Kendra retrieves the relevant excerpts and an LLM turns them into a contextual response.
Conversation with memory
LangChain maintains context, so a follow-up question is not treated as an unrelated one.
70% less search time
Retrieval response times halved, and manual searching dropped by 70%.
The stack
What it runs on.
- Search index
- Amazon Kendra
- Connectors
- Amazon S3, SharePoint, Confluence, website
- Model layer
- LLM with retrieved context in the prompt
- Conversation
- LangChain contextual conversation memory
The road ahead
The documentation is answerable. Next is what else is.
With connectors and an index in place, extending coverage to further sources or document classes is configuration-and the same retrieval pattern applies to the next knowledge base rather than needing a new one.
- 01
More document classes indexed
- 02
Deeper troubleshooting workflows
- 03
Field-device access for technicians
Talk to us
Bring one process. Leave with an outcome architecture.
30 minutes on one function, mapped to the architecture behind it. No slides.