AI War Stories: Real world learnings delivering enterprise AI solutions
summary
Companies often get 80% to a working prototype but stall before production because of poor use case choice, bad data and unmet accuracy expectations. Practical wins include a one‑week automation that replaces a 25-person claims team and a university subject‑rationalization POC built in under a week that becomes a PowerBI cost saver. Architecture uses Azure blobs, embeddings, an AI Search (Lucene) index and the RAG pattern with APIM, private endpoints and metadata tagging; most effort goes into data engineering, indexing and access controls.
AI-written summary · about
chapters (10)
- 0:00Welcome to AI war stories
- 6:45AI projects struggle in production
- 13:28Enterprise AI use cases
- 21:56Verbs test for AI selection
- 27:57Stakeholder accuracy expectations
- 35:26Explaining LLMs and RAG
- 41:09RAG pattern pitfalls
- 47:07Data ingestion for AI search
- 51:00Chat vs search in AI
- 54:33Key takeaways for AI production
