Multi-tenant Conversational Analytics
summary
Built a multi-tenant conversational analytics system for cash-management customers that actually queries BigQuery with row-level security, stores embeddings in PostgreSQL vector store, and generates SQL and plotting code with Gemini for SQL and plotting. Deployment uses Google Cloud Platform (Cloud Run + GitHub Actions) and a Chainlit Python interface; an agentic decision-tree controls state instead of free-form LangChain. It works end-to-end but hits high DB load and session-affinity issues, and wide denormalized tables simplify reliable SQL generation.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 5:07What is conversational analytics
- 9:02Multi-tenant conversational analytics explained
- 12:22Key requirements for multi-tenancy
- 16:58Tooling and architecture decisions
- 21:18Implementation challenges and solutions
- 23:25Future trends and closing
- 24:29Q&A
related talks
45mManaging Data & Analytics Projects Effectively
45mForget SQL! Create your own AI Data Analyst
36mImplementing Natural Language Analytics with Snowflake Cortex
37mBuilding a Data Platform from scratch
24mScalable GenAI in Production
10mFrom Blank Page to Prod: A Year of Building AI Agents at Carrefour
58mDemystifying GenAI: Build a ChatGPT App with Vector Store
19mLeveraging Cassandra to Provide Generative AI That Powers Enterprise Analytics