Integrate Offline Vector Search with ChromaDB

[Vivekgupta008]

Description

The current BabyNest agent implementation passes all structured data from the database into a single, large context string for the LLM. This approach is inefficient, not scalable, and can result in the LLM receiving irrelevant information, diluting the context and leading to less precise answers.

This task is to replace the static context-injection method with a dynamic, offline-first vector search system powered by ChromaDB. This will make the agent's responses more accurate and context-aware by retrieving only the information that is semantically relevant to a user's query, while maintaining full offline capabilities.

Acceptance Criteria

• All structured data (appointments, symptoms, weight logs, etc.) from the SQLite database should be converted into embeddings and stored in a persistent ChromaDB collection.
• The agent's main workflow must be updated to query this vector store for relevant context before calling the LLM.
• The solution must be fully testable with a dedicated test script.