Built a RAG-powered support assistant that answers docs-based questions, classifies issues, and creates cleaner engineering tickets.
Challenge
A B2B SaaS team had a fast-growing support queue filled with repeated product questions, billing confusion, implementation issues, and bug reports. Docs existed, but customers did not find the right page. Engineers received vague tickets with missing reproduction steps.
Workflow approach
We created a RAG support agent trained on documentation, release notes, pricing rules, and known issues. The agent answers with citations, detects likely bugs, asks targeted reproduction questions, and creates tickets with environment data and clear severity. The system also reports content gaps when users ask questions the docs cannot answer.
Observed outcome
The assistant resolved 57% of support conversations without human handoff and cut first response time by 46%. Engineering received cleaner bug reports, duplicate tickets dropped 35%, and the docs team gained a prioritized list of missing articles.
Outcome metrics
- Tickets resolved by AI
- 57%
- Faster first response
- 46%
- Citation accuracy
- 91%
- Fewer duplicate bugs
- 35%
Systems involved
- Claude
- Pinecone
- LangChain
- Intercom API
- Linear API
- TypeScript