Development
RAG assistant grounds support answers in current documentation
A B2B SaaS support team stopped answering from a stale wiki and started answering from source-cited RAG results.
+34%
Deflection
0
Hallucinated facts
−41%
Median answer time
Challenge
Support agents kept quoting a wiki that had drifted from product reality. Escalations spiked whenever the wiki was wrong.
Solution
A RAG assistant indexes product docs, release notes, and known-issues, retrieves per-question, and cites source URLs. Weekly eval on a growing question set.
Stack
- TypeScript
- pgvector
- Postgres
- OpenAI compatible LLM
- Zendesk
