Building a Conversational Knowledge Base
summary
Built a conversational knowledge base that indexes robot docs and specs and returns focused answers using retrieval-augmented generation with embeddings and a vector DB, document parsing via Dockling document parsing, and tokenization via Vault for PII. Pipeline runs on Nomad with Azure blob storage and OpenWebUI for embeddings and agents. Demo works end to end but live upload/processing is flaky and OCR/low‑quality PDFs produce garbage in garbage out, so manual cleanup and a glossary materially improve accuracy.
AI-written summary · about
chapters (7)
- 0:00Conversational knowledge base intro
- 3:10Domain knowledge access challenges
- 9:53Retrieval augmented generation
- 14:36Knowledge base pipeline
- 23:19Fine-tuning not the solution
- 26:52Project lessons learned
- 28:40Resources and next steps
related talks
45mManaging Data & Analytics Projects Effectively
14mBuilding RAG at scale with Semantic Kernel and DataStax Astra DB
20mBuilding AI Applications with Azure Cosmos DB JavaScript SDK
12mBuilding Generative AI Apps with Azure Cosmos DB
37mBuilding an Enterprise Knowledge Management System using Hybrid Search in Azure Cosmos DB
28mBuild an AI Research Agent
01h 17mGraph Database and Retrieval Augmented Generation
15mGraphRAG-Powered AI Agents: Transforming Corporate Metadata into Intelligent Knowledge