Streaming Data to RAG Developer Example

Why it matters

A live source helps an answer only once it is searchable. This example shows the gap between a transcript arriving and becoming retrievable, in its own interface.

What to practise

Deploy the Context-Aware RAG services (Milvus, Neo4j, ingestion and retrieval), then the replay profile: the file-replay service, the Holoscan software-defined radio, the Parakeet ASR NIM and the NeMo Agent Toolkit UI. Watch transcripts appear in the UI, and on the history page watch each one change state once its batch is processed. Find chat.params.batch_size and top_k in config.yaml, then ask questions bounded by channel and time window, such as the last ten minutes on one stream.

How it fits

This is 6.5.3’s streaming ingestion; compare the batch delay with 6.5.3’s searchable progress, and the time-window questions with its use of source time to resolve stale answers. Its Context-Aware RAG and Agent Toolkit UI come from streaming-data-to-rag-dev development branches, and the UI image keeps the earlier agentiq-ui name.