One agent, a knowledge base that grew up.
TracE is a single RAG copilot — the story is its grounding. Step through how the knowledge base moved from a curated PDF corpus in the POC, to a daily-indexed Confluence connector across three spaces, to the production mentor that cites its source on every answer and abstains when the documentation isn’t there.
TracE in production — the mentor copilot
TracE in live production for the Service Desk team — 75 technicians, ~40 daily active. A friendly-mentor persona breaks complex processes into clear steps, defines acronyms, and cites its source on every answer. When the documentation isn’t there, it says so and guides the technician on next steps rather than guessing. The deployment pipeline built here was replicated for DJ and Oscar.
Hover or tap an agent above to see its pipeline and what it returns for this bid.
◆TracE— Production Service Desk Copilot
The production agent. Every response follows a fixed structure — a short summary, the clear answer as numbered steps, important reminders, and a link to the source article. The persona is a supportive team lead sitting next to the technician. Cite-or-abstain is absolute: when the knowledge isn’t documented, TracE flags the gap and guides next steps instead of inventing an answer — the deterministic, structured-knowledge-first doctrine the rest of the portfolio is built on.
- SourceConfluence daily sync (3 spaces) · every query — primaryauthoritative, near real-time
- KnowledgeGlossaries & abbreviation guides · term resolution
- FallbackLegacy PDF corpus · specialized / legacy content
- ToolKnowledge-gap handling · when nothing is documentedsays so + guides next steps — never guesses
Why it matters · Consistent, cited, cite-or-abstain answers let L1/L2 techs resolve more themselves with confidence — and set the response contract every later MillerKnoll copilot follows.