Implement Typesense support for RAG in the AI module

I implemented an AI Agent that performs
RAG (Retrieval-Augmented Generation) on top of
Typesense’s semantic search.

When a user enters a query (as a prompt), the main agent doesn’t just run it directly.
Instead, it first calls a sub-agent whose job is to determine
which Typesense collection is most relevant to the query.

Once the sub-agent identifies the best collection, the main agent uses that collection to perform a
semantic search in Typesense (following
Typesense’s semantic search workflow).
The retrieved context is then fed back into the AI to generate an enriched, context-aware answer—completing the RAG cycle.

• Dynamic routing: It automatically selects the right collection based on intent.
• Scalability: No hardcoded mapping; adding new collections doesn’t require code changes.
• Accuracy: Queries hit the most relevant data, improving retrieval quality and answers.

• User enters a prompt.
• Main agent calls the sub-agent to pick the most relevant Typesense collection.
• Main agent runs Typesense semantic_search against that collection.
• Retrieved passages are injected into the LLM as context (RAG) to produce the final answer.