This image shows a multimodal retrieval and agent workflow that moves from enterprise video transcripts to database-powered agent responses for end users.

The workflow has the following components and handoffs:

Content source and transcript preparation.
The Video platform serves as an enterprise video library and provides source content.
Transcripts in SubRip subtitle (SRT) and Web Video Text Tracks (VTT), with metadata, move from transcript preparation to the ingestion pipeline.
The ingestion pipeline parses and normalizes transcript and metadata inputs, then outputs comma-separated values (CSV) data.
Database indexing and tool access.
Oracle Autonomous AI Database 26ai receives CSV-derived data and stores tables, embeddings, and a vector index.
Autonomous AI Database MCP Server connects to database content and exposes Select AI Agent tools.
Agent orchestration and user delivery.
runs a supervisory agent that invokes available tools.
The custom UI, shown with Streamlit as an example, receives agent output and presents responses.
End users submit requests through the Custom UI and receive results returned by the workflow.