AI Integration
Timeline2 weeks
StackRAG · API · Graphiti · Obsidian
01 / Challenge
The company had hundreds of pages of documentation. Managers answered the same questions for hours, new employees drowned in the knowledge base.
02 / Approach
We built a RAG assistant over the knowledge base: documents are indexed, the assistant streams answers and links the source of every claim. Accuracy is held by evals on real questions, hallucinations are cut by mandatory citations.
03 / Steps
01
Knowledge base
cleaned and indexed the documents
02
RAG pipeline
retrieval + generation with source links
03
Evals
runs on real questions, accuracy tuning
04
Embedding
a site widget and a bot for the team
04 / Result
1.2sanswer
2.1Mtokens/mo
P95stable
An answer in 1.2 seconds instead of hours of waiting. 2.1 million tokens in production monthly, stable P95. Managers are unloaded, newcomers find answers themselves.