Building a Meeting Intelligence Platform to Tackle Context-Switching

The author built an enterprise meeting intelligence platform to address the problem of scattered context across various tools like Teams, Jira, and Snowflake. The platform ingests meeting transcripts, Jira tickets, and data schemas to provide grounded answers to user questions.

Details

The author, who works on an analytics team, faced the challenge of having relevant information scattered across various tools like Microsoft Teams, Jira, and Snowflake. To address this, they built an enterprise meeting intelligence platform that ingests meeting transcripts, Jira tickets, and Snowflake data schemas to provide grounded answers to user questions. The platform uses a modular pipeline architecture with separate steps for ingestion, chunking, embedding, retrieval, and generation. It leverages ChromaDB for local vector storage and Snowflake Cortex for LLM-based answer generation, keeping sensitive enterprise data within the Snowflake environment. The author shares key lessons learned, such as the importance of testing on real-world data, handling metadata properly, and balancing automation with user control. Future posts will cover topics like parsing messy Teams transcripts, the hybrid ChromaDB and Snowflake architecture, and the capabilities and limitations of Snowflake Cortex for enterprise RAG.