Knowledge
Use what the team knows.
Confluence provides the knowledge base. Retrieval brings relevant information into the assistant’s response process.
Applied AI / Fintech support
Put the knowledge base
to work in support.
A support assistant connecting the information in Confluence with the questions arriving in Zendesk. Routine answers, with a route to a person.
Knowledge → supportHuman escalation
Knowledge sourceConfluenceThe team’s existing support knowledge
Retrieve + respondAI support assistantRelevant information brought into the response
Support workflowZendesk + the support teamRoutine answers with a path to a person
At a glance
- Client
- A fintech scale-up in the EU. Not named.
- Constraint
- A lean support team and growing ticket volumes.
- Contribution
- A retrieval-based assistant connecting Confluence knowledge to Zendesk, with human escalation.
- Result
- The assistant handled repetitive queries, leaving the team more time for complex issues.
01 / The brief
More questions.
The same support team.
Growing ticket volumes were overwhelming a lean support team. Repetitive enquiries competed with the issues that needed closer attention.
The team already had two essential pieces in place: Zendesk for support and Confluence for knowledge. The brief was to bring that information into the answering process.
Human escalation was part of the solution. Handling routine questions had to sit within a support workflow that could still involve the team.
02 / What we built
Connect the question
to the knowledge.
Devopsly built a retrieval-augmented generation assistant and integrated it with the client’s existing support tools.
Assistant
Give the model context.
The assistant uses a language model to respond with retrieved information as context. This approach is known as retrieval-augmented generation, or RAG.
Workflow
Keep a human handoff.
The Zendesk integration places the assistant in the support workflow, with human-in-the-loop escalation for the team to take over.
Delivery included OpenAI, Python, LangChain and AWS, alongside the Zendesk and Confluence integrations.
03 / The result
Routine questions handled.
Room for complex support.
The assistant handled repetitive queries while the support team focused on more complex issues.
The delivered system connects the knowledge source, the answering capability and the ticketing workflow. Human escalation remains part of that workflow.
A qualitative result. The client is not named and no measured figures are published for this project.
A similar challenge?
Start with the questions
your team sees every day.
A useful first conversation covers recurring ticket types, the state of your knowledge base and what should always reach a person.
Discuss your support workflowMore selected work
Different problems. The same care.
Card issuance product
The ledger, cloud and release pipelines behind a debit card.
Read the case studyWinona mobile app
Wine collection software with offline capability.
Read the case studyHR and payroll assistant
Permission-aware answers to HR and pay questions.
Read the case study