Build and Deploy the Video Transcript RAG Solution

Build the transcript ingestion, vector retrieval, agent orchestration, and user interface flow for this video transcript solution playbook.

To build and deploy the Video Transcript RAG solution, follow these steps:

1. Prepare transcripts for loading by converting SRT or VTT files into a CSV format that the database load process can ingest.

   Collect transcript files for each video, with optional locale suffixes, and run a parser that outputs a transcript CSV.

   Use these columns in the CSV: video_id , title , transcript_index , locale , timestamp_start , timestamp_end , and transcript_text .

2. Provision Autonomous AI Database 26ai with Terraform, if you use the infrastructure-as-code path.

   Configure Terraform to create the schema, load transcripts into a staging table, generate embeddings, build a vector index, and create a Select AI Agent retrieval tool.

   Enable MCP support through the Autonomous AI Database freeform tag adb$feature with value mcp_server , and capture the mcp_server_url output.

3. Initialize schema objects and load transcript data with SQL scripts when you run the SQL path directly or validate Terraform execution.

   Create tables and indexes, load transcript CSV data into a staging table, and generate baseline embeddings for the primary locale.

   Create the Select AI Agent retrieval tool and the supporting search function.

4. Expose retrieval through MCP by publishing search_videos_tool as a Select AI Agent tool.

   Configure the tool to return JSON fields for video_id , title , timestamp_start , transcript_text , and vector_distance .

5. Configure Oracle AI Agent Studio to orchestrate question handling, retrieval, and grounded summarization.

   Define workflow behavior to accept a question, call search_videos_tool with query and locale, and return a concise answer with citations.

   Include a list of playable video moments with video ID, title, start timestamp, and snippet.

6. Run the sample user interface and verify that responses include timestamped playback links.

   The demo Streamlit interface sends questions to Oracle AI Agent Studio and renders summary output with embedded playback behavior.

   Set environment variables for database connectivity, agent access, and playback URL construction.

   If Oracle AI Agent Studio isn't available, call database retrieval directly from the UI to validate end-to-end vector retrieval.

You now have an end-to-end implementation for ingesting transcripts, retrieving relevant video moments, and returning grounded answers with timecoded playback.