A journey from static documentation to generative AI
summary
Large enterprise documentation becomes a conversational RAG system that uses GPT-4 and text-embedding-large-free on Azure, ingesting Confluence, SharePoint, Jira and GitHub via a modular scraper into blob storage, indexers and a vector DB. A seven‑engineer team proves a 4‑day POC then months of hardening with guardrails, RBAC, version backups, two‑tier cost throttling and observability. A 40–50 user beta drives token-cost sizing and automated RAC scoring (relevance, accuracy, completeness); results show measurable reduction in search time and steadily improving answer quality.
AI-written summary · about
chapters (18)
- 0:00Introduction
- 1:23Outdated documentation problem
- 3:07Generative AI solution
- 6:42Static docs attempt
- 7:44Knowledge hub shortcomings
- 9:08Dynamic knowledge docs
- 10:53Trustworthy chatbot challenges
- 11:37Retrieval augmented generation
- 15:53Answer reliability and hallucinations
- 17:38LLM cost and security
- 20:45Chatbot answer quality
- 23:17Observability and pipeline
- 26:49Document sources and cleaning
- 28:59Production and version control
- 30:01RAG solution learnings
- 31:06Living documentation future
- 33:10Impact and next steps
- 35:14Q&A
related talks
47mRevamping the Document Automation Workflow with Generative AI
22mLearning to Apply Generative AI at Enterprise Level
23mDocumentation in the Age of AI: Why Docs Are a Product
28mImproving responses in generative AI applications
39mDocs AI Tooling is Better (and Better for Us) than You Think
36mGENAI-Powered chatbot to increase team efficiency
15mHarnessing the Power of Generative AI for Intelligent Applications
40mJonathon Wright: Generative AI: Turning Testers into Superhero’s